The Rank-Two Split of $M_4(\mathbb{C})$
Introduction
The rank-two idempotent $\Pi=P_+$ cuts $M_4(\mathbb{C})$ into four corners of dimension four each, and the diagonal corners are two copies of $M_2(\mathbb{C})$. The split is the balanced split $M_4\supset M_2\oplus M_2$; the off-diagonal corners are the two mass blocks, and the block-diagonal subalgebra is not an ideal. The split is the chiral block decomposition of The Two Chiralities of $M_4(\mathbb{C})$, read as a Peirce decomposition.
The article states the rank-two idempotent, proves the corner dimensions $4+4+4+4$, identifies the two diagonal corners with $M_2(\mathbb{C})$, describes the two carrier modules and the mass block, gives the two-sided Peirce table and the multiplication table, and explains why the split is a vector-space decomposition and a subalgebra inclusion, not a direct sum of algebras.
The rank-one split is The Peirce Decomposition of $M_4(\mathbb{C})$; the chirality is The Two Chiralities of $M_4(\mathbb{C})$; the mass block and its polar decomposition are The Mass Block and the Chiral Polar Decomposition of $M_4(\mathbb{C})$; the two $\mathfrak{su}_2$'s are The Peirce $\mathfrak{su}_2$ and the Chiral $\mathfrak{su}_2\times\mathfrak{su}_2$ of $M_4(\mathbb{C})$. The general recursion is The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$.
The Rank-Two Idempotent
Definition. The rank-two idempotent of the root is
$$ \Pi=P_+=\frac12(I+\Gamma)=P_1+P_3, $$
and its complement is
$$ Q=P_-=\frac12(I-\Gamma)=P_2+P_4 . $$
Proposition. The elements $\Pi,Q$ are idempotent, orthogonal and complete:
$$ \Pi^2=\Pi,\qquad Q^2=Q,\qquad \Pi Q=Q\Pi=0,\qquad \Pi+Q=I. $$
Each has rank $2$ and trace $2$. Their images are the two chiralities:
$$ \operatorname{im}\Pi=\mathbb{C}^2_+,\qquad \operatorname{im}Q=\mathbb{C}^2_- . $$
Proof. $\Pi$ and $Q$ are the spectral projectors of the involution $\Gamma$ for the eigenvalues $+1$ and $-1$. The algebraic identities follow from $\Gamma^2=I$; the ranks are the dimensions of the eigenspaces; the traces are the ranks.
The standard model. For $\Gamma_0=\operatorname{diag}(1,-1,1,-1)$,
$$ \Pi_0=\operatorname{diag}(1,0,1,0)=E_{11}+E_{33}, \qquad Q_0=\operatorname{diag}(0,1,0,1)=E_{22}+E_{44}. $$
In this frame the two chiralities are the coordinate planes $\operatorname{span}\{e_1,e_3\}$ and $\operatorname{span}\{e_2,e_4\}$.
The Four Corners
Definition. The rank-two Peirce corners are
$$ \mathcal{A}=\Pi M_4\Pi,\qquad \mathcal{B}=\Pi M_4Q,\qquad \mathcal{C}=QM_4\Pi,\qquad \mathcal{D}=QM_4Q . $$
Theorem. The algebra is the direct sum
$$ M_4(\mathbb{C})=\mathcal{A}\oplus\mathcal{B}\oplus\mathcal{C}\oplus\mathcal{D}, $$
with
$$ \dim_\mathbb{C}\mathcal{A}=4,\qquad \dim_\mathbb{C}\mathcal{B}=4,\qquad \dim_\mathbb{C}\mathcal{C}=4,\qquad \dim_\mathbb{C}\mathcal{D}=4. $$
The diagonal corners are full matrix algebras,
$$ \mathcal{A}\cong M_2(\mathbb{C}),\qquad \mathcal{D}\cong M_2(\mathbb{C}), $$
and the off-diagonal corners are the spaces of maps between the two chiralities,
$$ \mathcal{B}\cong\operatorname{Hom}(\mathbb{C}^2_-,\mathbb{C}^2_+)\cong M_2(\mathbb{C}), \qquad \mathcal{C}\cong\operatorname{Hom}(\mathbb{C}^2_+,\mathbb{C}^2_-)\cong M_2(\mathbb{C}). $$
Proof. Choose a basis adapted to $\mathbb{C}^4=\mathbb{C}^2_+\oplus\mathbb{C}^2_-$, with the first two vectors in $\mathbb{C}^2_+$ and the last two in $\mathbb{C}^2_-$. In that basis
$$ \Pi=\begin{pmatrix}I_2&0\\0&0\end{pmatrix}, \qquad Q=\begin{pmatrix}0&0\\0&I_2\end{pmatrix}, $$
and the four corners are the four $2\times2$ blocks of a block matrix. Each block has four entries, so each corner has dimension four. The diagonal blocks are closed under multiplication and are full matrix algebras with units $\Pi$ and $Q$; the off-diagonal blocks are the spaces of linear maps between the two planes.
The compressed units. The unit of $\mathcal{A}$ is $\Pi$, not $I$; the unit of $\mathcal{D}$ is $Q$. For an element $X$, the trace of its compression to $\mathcal{A}$ is $\operatorname{tr}(\Pi X\Pi)$, the trace of the $2\times2$ block, which differs from $\operatorname{tr}X$ in general; on $\mathcal{A}$ itself the two agree. The two diagonal corners are matrix algebras with their own units, and the embedding $\mathcal{A}\oplus\mathcal{D}\subset M_4$ is unital with unit $\Pi+Q=I$, while neither summand contains $I$.
The block form. A general element is
$$ A=\begin{pmatrix}A_{++}&A_{+-}\\A_{-+}&A_{--}\end{pmatrix}, \qquad A_{++}\in\mathcal{A},\quad A_{+-}\in\mathcal{B},\quad A_{-+}\in\mathcal{C},\quad A_{--}\in\mathcal{D}. $$
The product is
$$ \begin{pmatrix}A&B\\C&D\end{pmatrix} \begin{pmatrix}A'&B'\\C'&D'\end{pmatrix} = \begin{pmatrix} AA'+BC'&AB'+BD'\\ CA'+DC'&CB'+DD' \end{pmatrix}. $$
The products $BC'\subseteq\mathcal{A}$ and $CB'\subseteq\mathcal{D}$ are the cross terms; they are the reason the split is not an algebra direct sum.
The Two Carrier Modules
The two carrier spaces. The idempotent $\Pi$ defines the left module
$$ L_+=M_4\Pi=\{X\Pi:X\in M_4\}, $$
of complex dimension $8$, and $Q$ defines
$$ L_-=M_4Q=\{XQ:X\in M_4\}, $$
also of dimension $8$. These are the two algebraic carrier modules of the split. The first is the set of matrices that kill the second chirality, equivalently the space of maps $\mathbb{C}^2_+\to\mathbb{C}^4$; the second is the set that kills the first chirality, equivalently the space of maps $\mathbb{C}^2_-\to\mathbb{C}^4$.
Proposition. As a left module over the block-diagonal algebra $\mathcal{A}\oplus\mathcal{D}$,
$$ L_+\cong\mathcal{A}\oplus\mathcal{C},\qquad L_-\cong\mathcal{B}\oplus\mathcal{D}. $$
Proof. A matrix in $L_+$ has zero second block-column; its first block-column has a top part in $\mathcal{A}=\Pi M\Pi$ and a bottom part in $\mathcal{C}=QM\Pi$. The left action of $\mathcal{A}\oplus\mathcal{D}$ preserves the row decomposition and acts on the two parts by $\mathcal{A}$ and $\mathcal{D}$ respectively. The same argument with the second block-column gives $L_-\cong\mathcal{B}\oplus\mathcal{D}$.
The two carrier spaces of the principal $\mathfrak{sl}_2$. The principal $\mathfrak{sl}_2$ of the frame has one carrier, $\mathbb{C}^4=\operatorname{sym}^3(\mathbb{C}^2)$. The rank-two split has two carrier spaces, $\mathbb{C}^2_+$ and $\mathbb{C}^2_-$, one per chirality. They are the modules of the two $M_2$ factors:
$$ \mathcal{A}\ \text{acts on}\ \mathbb{C}^2_+, \qquad \mathcal{D}\ \text{acts on}\ \mathbb{C}^2_- . $$
The maps in $\mathcal{B}$ go from $\mathbb{C}^2_-$ to $\mathbb{C}^2_+$; the maps in $\mathcal{C}$ go from $\mathbb{C}^2_+$ to $\mathbb{C}^2_-$. The two carrier spaces are two-dimensional; the principal $\mathfrak{sl}_2$ carrier is four-dimensional. They are different objects with similar names.
The mass block. The corner
$$ \mathcal{B}=\Pi M_4Q=\operatorname{Hom}(\mathbb{C}^2_-,\mathbb{C}^2_+) $$
is the mass block. It is four-dimensional, and it is a bimodule: $\mathcal{A}$ acts on the left and $\mathcal{D}$ acts on the right. An element $M\in\mathcal{B}$ is a linear map
$$ M:\mathbb{C}^2_-\longrightarrow\mathbb{C}^2_+, $$
the mass operator of the split. The opposite corner $\mathcal{C}$ is the conjugate mass block, and the two together form the odd part of the algebra under $\operatorname{Ad}_\Gamma$. The polar decomposition of $M$ is The Mass Block and the Chiral Polar Decomposition of $M_4(\mathbb{C})$.
The Two-Sided Peirce Table
The four corners form a $2\times2$ table:
| $\Pi$ | $Q$ | |
|---|---|---|
| $\Pi$ | $\mathcal{A}\cong M_2(\mathbb{C})$ | $\mathcal{B}\cong M_2(\mathbb{C})$ |
| $Q$ | $\mathcal{C}\cong M_2(\mathbb{C})$ | $\mathcal{D}\cong M_2(\mathbb{C})$ |
The entry in row $\Pi$ and column $\Pi$ is the corner $\Pi M\Pi$; the entry in row $\Pi$ and column $Q$ is $\Pi MQ$; and so on. The table is the two-sided Peirce table of the pair $(\Pi,Q)$.
The multiplication table. The product of the corner in row $i$ and column $j$ with the corner in row $k$ and column $l$ is contained in the corner in row $i$ and column $l$ when $j=k$, and is zero otherwise:
$$ (\Pi_iM\Pi_j)(\Pi_kM\Pi_l)\subseteq\delta_{jk}\Pi_iM\Pi_l, $$
where $\Pi_1=\Pi$ and $\Pi_2=Q$. Written as a table:
| $\cdot$ | $\mathcal{A}$ | $\mathcal{B}$ | $\mathcal{C}$ | $\mathcal{D}$ |
|---|---|---|---|---|
| $\mathcal{A}$ | $\mathcal{A}$ | $\mathcal{B}$ | $\mathcal{C}$ | $0$ |
| $\mathcal{B}$ | $0$ | $0$ | $\mathcal{A}$ | $\mathcal{B}$ |
| $\mathcal{C}$ | $\mathcal{C}$ | $\mathcal{D}$ | $0$ | $0$ |
| $\mathcal{D}$ | $0$ | $0$ | $\mathcal{C}$ | $\mathcal{D}$ |
The refinement into lines. Each four-dimensional corner is the sum of four fine Peirce lines:
$$ \mathcal{A}=\mathcal{A}_{11}\oplus\mathcal{A}_{13}\oplus\mathcal{A}_{31}\oplus\mathcal{A}_{33}, $$
$$ \mathcal{B}=\mathcal{A}_{12}\oplus\mathcal{A}_{14}\oplus\mathcal{A}_{32}\oplus\mathcal{A}_{34}, $$
$$ \mathcal{C}=\mathcal{A}_{21}\oplus\mathcal{A}_{23}\oplus\mathcal{A}_{41}\oplus\mathcal{A}_{43}, $$
$$ \mathcal{D}=\mathcal{A}_{22}\oplus\mathcal{A}_{24}\oplus\mathcal{A}_{42}\oplus\mathcal{A}_{44}. $$
Each line is one-dimensional, and the sixteen lines are the full Peirce decomposition of The Peirce Decomposition of $M_4(\mathbb{C})$. The rank-two split groups the sixteen lines into four groups of four. It is a different grouping from the rank-one split, whose corners have dimensions $1,3,3,9$; neither grouping refines the other.
The Failure of the Algebra Direct Sum
The vector-space direct sum is not an algebra direct sum. The algebra is the direct sum of the four corners as vector spaces, but it is not the direct sum of $\mathcal{A}$ and $\mathcal{D}$ as algebras. The off-diagonal blocks $\mathcal{B}$ and $\mathcal{C}$ are nonzero, and the cross terms
$$ \mathcal{B}\mathcal{C}\subseteq\mathcal{A},\qquad \mathcal{C}\mathcal{B}\subseteq\mathcal{D} $$
return to the diagonal. If the off-diagonal blocks were zero, the algebra would be $\mathcal{A}\oplus\mathcal{D}$; they are not, and the algebra is larger.
The block-diagonal part is a subalgebra, not an ideal. The subspace
$$ \mathcal{A}\oplus\mathcal{D}=\left\{\begin{pmatrix}A&0\\0&D\end{pmatrix}\right\} $$
is closed under multiplication and contains the unit $I$. It is isomorphic to $M_2(\mathbb{C})\oplus M_2(\mathbb{C})$, the balanced subalgebra $M_2\oplus M_2$ of the recursion. It is not a two-sided ideal: multiplying a block-diagonal element by a general element produces an off-diagonal part. Indeed, $M_4(\mathbb{C})$ is simple, so it has no proper two-sided ideal at all.
The meaning of "contains". When the category says that $M_4$ contains $M_2\oplus M_2$, the word means the block-diagonal subalgebra, not an ideal and not a quotient. The inclusion is
$$ M_2(\mathbb{C})\oplus M_2(\mathbb{C})\longrightarrow M_4(\mathbb{C}),\qquad (A,D)\longmapsto\begin{pmatrix}A&0\\0&D\end{pmatrix}, $$
and the off-diagonal blocks complete the image to the whole algebra. The same convention applies to the other categories: $M_6$ contains $M_3\oplus M_3$ and $M_5$ contains $M_2\oplus M_3$ as block-diagonal subalgebras.
The trace and the determinant. The trace splits as
$$ \operatorname{tr}A=\operatorname{tr}A_{++}+\operatorname{tr}A_{--}, $$
a sum of the traces of the two diagonal blocks. The determinant is the determinant of the full block matrix; when $A_{--}$ is invertible,
$$ \det A=\det(A_{++}-A_{+-}A_{--}^{-1}A_{-+})\det A_{--}, $$
and the formula extends by continuity when $A_{--}$ is singular.
The mass block enters through the Schur complement $A_{+-}A_{--}^{-1}A_{-+}$, which is a map on $\mathbb{C}^2_+$. The determinant is not the product $\det A_{++}\det A_{--}$ unless the mass block is zero.
Comparison with the Rank-One Split and with $M_2$
The rank-one split. For the rank-one idempotent $\Pi=P_1$, the corners are
$$ 1+3+3+9, $$
with $\mathcal{A}\cong\mathbb{C}$, $\mathcal{D}\cong M_3$, $\mathcal{B}\cong\mathbb{C}^{1\times3}$ and $\mathcal{C}\cong\mathbb{C}^{3\times1}$. The rank-two split replaces the line and the three-space by two planes:
$$ 1+3+3+9\ \longrightarrow\ 4+4+4+4. $$
The rank-two split is the balanced split $M_2\oplus M_2$; the rank-one split is the unbalanced split $\mathbb{C}\oplus M_3$. The balanced split is the one adapted to the two chiralities, because both chiralities have the same dimension.
The biquaternion case. For $M_2(\mathbb{C})$, a rank-one idempotent $\Pi$ has complement $Q$ of rank one, and the split is
$$ 1+1+1+1. $$
There is no rank-two idempotent other than $I$, whose complement is $0$, so the balanced split $M_1\oplus M_1=\mathbb{C}\oplus\mathbb{C}$ is the only balanced split, and it is degenerate. The rank-two split $M_2\oplus M_2$ is a feature of $M_4$: it is the first case in which both chiralities are planes and both diagonal corners are non-scalar matrix algebras.
The recursion. The rank-two split is the case $p=2$ of the corner recursion
$$ M_n\supset M_p\oplus M_{n-p}, $$
and for $n=4$, $p=2$ gives $M_4\supset M_2\oplus M_2$. The rank-one split is the case $p=1$, $M_4\supset M_1\oplus M_3$. The two are the two nontrivial balanced-or-not splits of $M_4$; the third, $p=3$, is the complement of $p=1$ and gives the same dimensions in the opposite order.
Summary
The rank-two idempotent $\Pi=P_+$ and its complement $Q=P_-$ cut the algebra into four corners of dimension four each:
$$ M_4(\mathbb{C})=\mathcal{A}\oplus\mathcal{B}\oplus\mathcal{C}\oplus\mathcal{D}, \qquad 4+4+4+4=16, $$
with $\mathcal{A}\cong M_2(\mathbb{C})$, $\mathcal{D}\cong M_2(\mathbb{C})$, and $\mathcal{B},\mathcal{C}$ the mass blocks, each isomorphic to $M_2(\mathbb{C})$ as a vector space and to $\operatorname{Hom}(\mathbb{C}^2_-,\mathbb{C}^2_+)$ or its opposite as a bimodule. The diagonal part $\mathcal{A}\oplus\mathcal{D}$ is the block-diagonal subalgebra $M_2\oplus M_2$; it is not an ideal, and the cross terms $\mathcal{B}\mathcal{C}\subseteq\mathcal{A}$ and $\mathcal{C}\mathcal{B}\subseteq\mathcal{D}$ show that the split is not an algebra direct sum. The two algebraic carrier modules are $L_+=M_4\Pi\cong\mathcal{A}\oplus\mathcal{C}$ and $L_-=M_4Q\cong\mathcal{B}\oplus\mathcal{D}$, each of dimension eight; the mass block is the four-dimensional corner $\mathcal{B}$. The rank-two split is the balanced refinement of the rank-one split $1+3+3+9$ and the four-dimensional analogue of the biquaternion split $1+1+1+1$.
$$ \boxed{\ M_4=\mathcal{A}\oplus\mathcal{B}\oplus\mathcal{C}\oplus\mathcal{D},\qquad 4+4+4+4=16,\qquad \mathcal{A}\cong\mathcal{D}\cong M_2(\mathbb{C}),\qquad \mathcal{B}\cong\mathcal{C}\cong M_2(\mathbb{C}).\ } $$
| Object | Dimension | Algebra or module |
|---|---|---|
| $\Pi=P_+$ | rank $2$ | idempotent |
| $Q=P_-$ | rank $2$ | complement |
| $\mathcal{A}=\Pi M\Pi$ | $4$ | $M_2(\mathbb{C})$ |
| $\mathcal{D}=QMQ$ | $4$ | $M_2(\mathbb{C})$ |
| $\mathcal{B}=\Pi MQ$ | $4$ | mass block, $\operatorname{Hom}(\mathbb{C}^2_-,\mathbb{C}^2_+)$ |
| $\mathcal{C}=Q M\Pi$ | $4$ | conjugate mass block |
| $\mathcal{A}\oplus\mathcal{D}$ | $8$ | block-diagonal $M_2\oplus M_2$ |
| $\mathcal{B}\oplus\mathcal{C}$ | $8$ | odd part |
| $L_+=M_4\Pi$ | $8$ | $\mathcal{A}\oplus\mathcal{C}$ |
| $L_-=M_4Q$ | $8$ | $\mathcal{B}\oplus\mathcal{D}$ |
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\Pi=P_+$ | the rank-two idempotent, $\frac12(I+\Gamma)$ |
| $Q=P_-$ | the complement, $\frac12(I-\Gamma)$ |
| $\mathcal{A}=\Pi M_4\Pi$ | the left chiral corner, $\cong M_2(\mathbb{C})$ |
| $\mathcal{D}=QM_4Q$ | the right chiral corner, $\cong M_2(\mathbb{C})$ |
| $\mathcal{B}=\Pi M_4Q$ | the mass block |
| $\mathcal{C}=QM_4\Pi$ | the conjugate mass block |
| $L_+=M_4\Pi$, $L_-=M_4Q$ | the two algebraic carrier modules |
| $\mathbb{C}^2_\pm$ | the two carrier spaces |
| $\mathcal{A}_{kl}$ | the fine Peirce lines inside the corners |
Further Reading
- $M_4(\mathbb{C})$ Axis-Free Representation, for the coefficient basis, the four left matrices and the transpose as the quaternion conjugation.
- The Two-Sided Operators Method to Build $M_4(\mathbb{C})$, for the matrix units, the sixteen two-sided operators and the caveat on the two bases.
- Richard S. Pierce, Associative Algebras (Springer, 1982), for Peirce decompositions relative to a pair of orthogonal idempotents and for corner algebras.
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge, 2013), for block matrices, Schur complements and the determinant.
- The Peirce Decomposition of $M_4(\mathbb{C})$, for the rank-one split and the sixteen fine lines.
- The Two Chiralities of $M_4(\mathbb{C})$, for the involution $\Gamma$ and the chiral blocks.
- The Mass Block and the Chiral Polar Decomposition of $M_4(\mathbb{C})$, for the mass operator and its polar decomposition.
- The Peirce $\mathfrak{su}_2$ and the Chiral $\mathfrak{su}_2\times\mathfrak{su}_2$ of $M_4(\mathbb{C})$, for the two $M_2$ factors and their Lie algebras.
- The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$, for the general recursion $M_n\supset M_p\oplus M_{n-p}$.
- Peirce Decompositions of the Biquaternion Algebra, for the biquaternion split $1+1+1+1$.