The Peirce Decomposition and the Corner Recursion of $M_4(\mathbb{C})$

Introduction

The corner $EM_4E$ of a rank-$k$ idempotent is a copy of $M_k(\mathbb{C})$ with unit $E$, and the whole algebra is rebuilt from the corner, its complement and the two rectangular blocks: $16=k^{2}+(4-k)^{2}+2k(4-k)$, with the trace adding over the two corners and the determinant obeying the Schur formula.

The article identifies the corner with a matrix algebra, proves the block reassembly and the multiplication rules, writes the trace and the determinant in Peirce coordinates with the Schur complement, works the two recursions at $k=2$ and at $k=1$, tabulates the Peirce basis, and reads the recursion at $k=2$ as the Kronecker factorisation $M_4\cong M_2\otimes M_2$.

It assumes The Peirce Decomposition and the Two Halves at $n=4$ for the spaces and The Kronecker Factorisation $M_4(\mathbb{C})\cong M_2(\mathbb{C})\otimes M_2(\mathbb{C})$ for the tensor reading. The model at order two is The Peirce Decomposition and the Corner Recursion of $M_2(\mathbb{C})$.

The Corner Is a Matrix Algebra

Theorem. For an idempotent $E$ of rank $k$ the corner $EM_4E$, with the multiplication of the algebra and the unit $E$, is isomorphic to $M_k(\mathbb{C})$.

Proof. Conjugate so that $E=\operatorname{diag}(1,\dots,1,0,\dots,0)$. Then $EME$ consists of the matrices with arbitrary $k\times k$ block in the top left and zeros elsewhere, and the map that keeps that block is an algebra isomorphism onto $M_k(\mathbb{C})$, carrying $E$ to $\mathrm I_k$.

Corollary. The complement $(1-E)M(1-E)$ is a copy of $M_{4-k}(\mathbb{C})$ with unit $1-E$, and the two corners are independent matrix algebras.

Proof. The same argument on the other block.

Corollary. $E$ is the identity of the corner and zero on the complement; an element $X$ is in the corner exactly when $EXE=X$.

Proof. $E(EAE)E=EAE$ and $E(1-E)M(1-E)E=0$.

The idempotent as a projection. The corner is the image of the projection $X\mapsto EXE$, and the complement is the image of $X\mapsto(1-E)X(1-E)$.

Proof. Both maps are idempotent linear maps with the images named.

The Block Reassembly

Theorem. $M_4=V_{11}\oplus V_{10}\oplus V_{01}\oplus V_{00}$ as vector spaces, and in the corresponding block notation $$ X=\begin{pmatrix}A&B\\ C&D\end{pmatrix},\qquad A\in M_k,\ B\in M_{k\times(4-k)},\ C\in M_{(4-k)\times k},\ D\in M_{4-k}, $$ the product is the block product and the dimensions are $k^{2}+k(4-k)+k(4-k)+(4-k)^{2}=16$.

Proof. Every matrix has a unique block decomposition, and the block product rule is the definition of the matrix product.

Corollary. The algebra is rebuilt from the two corner algebras and the two rectangular bimodules, and the multiplication of the pieces is the table $V_{ij}V_{kl}\subseteq\delta_{jk}V_{il}$.

Proof. The block product: the $B$ block multiplies into the $C$ block and the $C$ block multiplies into the $B$ block, which are the only two nonvanishing mixed products.

The recursion in words. $M_4$ is $M_k$ and $M_{4-k}$ glued by two rectangular blocks, and each gluing block is simultaneously a left module over one corner and a right module over the other.

Proof. $B$ is acted on by $A$ on the left and by $D$ on the right, and $C$ the other way round.

Corollary. The recursion terminates at matrices of size one, which are the scalars, so the algebra is built from the scalars by four gluing steps.

Proof. Repeated bisection of the index set reaches singletons.

Trace and Determinant in Peirce Coordinates

Proposition. $\operatorname{tr}X=\operatorname{tr}A+\operatorname{tr}D$; the two mixed blocks do not contribute.

Proof. The diagonal of $X$ is the diagonal of $A$ followed by the diagonal of $D$.

Corollary. The trace splits over the corners, and the traceless part is the direct sum of the traceless parts of the two corners together with the two mixed blocks.

Proof. Subtract the two traces.

Theorem (Schur). If $A$ is invertible then $$ \det X=\det A\cdot\det\bigl(D-CA^{-1}B\bigr), $$ and the identity holds as a polynomial identity for all $X$.

Proof. The block factorisation $$ \begin{pmatrix}A&B\\ C&D\end{pmatrix}=\begin{pmatrix}\mathrm I&0\\ CA^{-1}&\mathrm I\end{pmatrix}\begin{pmatrix}A&0\\0&D-CA^{-1}B\end{pmatrix}\begin{pmatrix}\mathrm I&A^{-1}B\\0&\mathrm I\end{pmatrix} $$ has determinant $\det A\det(D-CA^{-1}B)$, since the two triangular factors have determinant one. Both sides are polynomials in the entries, and they agree on the open dense set where $A$ is invertible, hence everywhere.

Corollary. For a block triangular matrix, $C=0$ or $B=0$, the determinant is $\det A\det D$, and the determinant is multiplicative over the recursion.

Proof. The Schur complement reduces to $D$ or to $D-CA^{-1}B$ with one of the two mixed blocks zero.

Corollary. $\det X$ is the product of the corner determinant and the determinant of the Schur complement, which is the "effective" complement after eliminating the corner.

Proof. The theorem.

The rank reading. The rank of $X$ is at least the rank of $A$ plus the rank of the Schur complement, with equality when $A$ has full rank in its block.

Proof. The factorisation above exhibits $X$ as the product of three matrices of ranks $\operatorname{rank}A$, $\operatorname{rank}A+\operatorname{rank}(D-CA^{-1}B)$ and $4$, so the rank of $X$ is that of the middle factor.

The Two Steps

The step $k=2$. $M_4=M_2\oplus M_2\oplus M_{2\times2}\oplus M_{2\times2}$ with the dimensions $4+4+4+4$, and the corner and the complement are both copies of $M_2$.

Proof. The dimension formula at $k=2$.

Corollary. At $k=2$ the two corners are the two factors of the Kronecker decomposition, and the mixed blocks are the off-diagonal parts of the tensor product.

Proof. With $E=\operatorname{diag}(1,0)\otimes\mathrm I_2$ the corner is $\operatorname{diag}(1,0)M_2\operatorname{diag}(1,0)\otimes M_2\cong\mathbb{C}\otimes M_2\cong M_2$.

The step $k=1$. $M_4=\mathbb{C}\oplus M_3\oplus\mathbb{C}^{3}\oplus(\mathbb{C}^{3})^{\ast}$ with the dimensions $1+9+3+3$, and the corner is the scalars while the complement is $M_3$.

Proof. The dimension formula at $k=1$.

Corollary. The step $k=1$ exhibits $M_4$ as $M_3$ with one row and one column adjoined, and the corner is the pivot of the elimination.

Proof. $B$ is a column, $C$ a row, and $A$ a scalar; the Schur complement is $D-CB/A$ with $A$ the pivot.

The two steps together. The recursion at $k=2$ splits the algebra into two halves of equal size, and the recursion at $k=1$ splits off one dimension at a time; the two are independent and neither refines the other.

Proof. The idempotents $\operatorname{diag}(1,1,0,0)$ and $\operatorname{diag}(1,0,0,0)$ are not nested in the sense of Peirce refinement, since the second is not in the corner of the first.

The Peirce Basis and Its Table

Definition. The Peirce basis at $E=\operatorname{diag}(1,1,0,0)$ is the sixteen matrix units, grouped as $E_{ij}$ with $i,j\in\{1,2\}$ in $V_{11}$, $i\in\{1,2\}$, $j\in\{3,4\}$ in $V_{10}$, $i\in\{3,4\}$, $j\in\{1,2\}$ in $V_{01}$ and $i,j\in\{3,4\}$ in $V_{00}$.

Proposition. The product table in the Peirce basis is the matrix-unit table $E_{ij}E_{kl}=\delta_{jk}E_{il}$, and it is block diagonal in the four groups.

Proof. The matrix-unit rule, read with the indices grouped.

Corollary. The four groups are closed under the product exactly in the pattern $V_{ij}V_{kl}\subseteq\delta_{jk}V_{il}$, and the Peirce basis is the matrix-unit basis read in blocks.

Proof. The index condition $\delta_{jk}$ says that the second index of the first factor matches the first index of the second, which is exactly the block adjacency condition.

The two-number table. In the four groups the products of $E_{ij}$ with $E_{kl}$ vanish unless $j=k$, and then produce the unit with the outer indices; the number of nonzero products is the number of triples $(i,j,l)$ with the indices in the right groups.

Proof. The table, counted.

Corollary. The Peirce basis is not the axis-free basis: the sixteen axis-free elements are signed permutations and the Peirce elements are matrix units, and the change between them is a unitary matrix.

Proof. The two bases are orthonormal for the Hilbert–Schmidt form up to the scale two, and they differ.

The Kronecker Reading

Theorem. At $k=2$ the recursion is the Kronecker factorisation: $M_4\cong M_2\otimes M_2$, where the first factor is generated by the corner and the complement and the second by the mixed blocks.

Proof. With $E=\operatorname{diag}(1,0)\otimes\mathrm I_2$ the corner is $\operatorname{diag}(1,0)M_2\operatorname{diag}(1,0)\otimes M_2$, the complement is $\operatorname{diag}(0,1)M_2\operatorname{diag}(0,1)\otimes M_2$, and the two mixed blocks are the off-diagonal parts of the first factor tensored with $M_2$.

Corollary. The two factors of the Kronecker decomposition are the two halves of the recursion, and the Peirce decomposition at $k=2$ is the block decomposition of the tensor product read along the first factor.

Proof. The tensor product of block matrices is the block matrix of the tensor products.

The contrast. At $k=1$ and at $k=3$ the recursion is not a tensor factorisation: the corner is $\mathbb{C}$ and the complement is $M_3$, and no tensor product of the two has dimension $16$ except the trivial one.

Proof. $1\cdot9=9\neq16$.

Corollary. The recursion is the general construction and the Kronecker factorisation is the case of the half rank, where the corner and the complement have the same size.

Proof. The identity $k^{2}+(4-k)^{2}+2k(4-k)=16$ is the dimension of a tensor product exactly when $k=4-k=2$.

Summary

The corner of a rank-$k$ idempotent is a copy of $M_k$ with unit $E$, and the algebra is rebuilt from the corner, the complement $M_{4-k}$ and the two rectangular blocks, with $16=k^{2}+(4-k)^{2}+2k(4-k)$. The trace adds over the two corners and the mixed blocks contribute nothing.

The determinant obeys the Schur formula $\det X=\det A\det(D-CA^{-1}B)$, which is an identity of polynomials and reduces to a product for block triangular matrices. The recursion at $k=2$ gives two copies of $M_2$ and two copies of $M_{2\times2}$, and at $k=1$ gives $\mathbb{C}$, $M_3$ and the column and row spaces.

The Peirce basis is the matrix-unit basis read in blocks, with the product table $E_{ij}E_{kl}=\delta_{jk}E_{il}$ and the block adjacency condition. At $k=2$ the recursion is exactly the Kronecker factorisation, and at the other ranks it is not, so the factorisation is the balanced case of a construction that works at every rank.

Summary of Notation

Symbol Meaning
$E$ idempotent of rank $k$
$EME$ the corner, a copy of $M_k$
$B,C$ the two rectangular blocks
$D-CA^{-1}B$ the Schur complement
$k^{2}+(4-k)^{2}+2k(4-k)$ the dimension count, equal to $16$
$M_2\otimes M_2$ the Kronecker reading at $k=2$

Further Reading