The Geometry of the Idempotents of $M_4(\mathbb{C})$

Introduction

The idempotents of rank $p$ in $M_4(\mathbb{C})$ form the homogeneous space $PGL_4/P(GL_p\times GL_{4-p})$, of complex dimension $2p(4-p)$; for $p=1$ this is an affine bundle of rank $3$ over $\mathbb{P}^3$, of dimension $6$. The fibre over a $p$-plane is the affine space of complements, which is contractible; the idempotents of rank $p$ are therefore homotopy equivalent to the Grassmannian $\operatorname{Gr}(p,4)$. The full flag variety $GL_4/B$ has dimension $6$; the partial flags and their incidence relations organise the complements, and the gauge torus is the fibre of the frame bundle over the root orbit.

The article describes the rank-one idempotents and their affine bundle, the general rank-$p$ homogeneous space, the tangent space and the rank stratification, the relation to the flag variety and the Grassmannians, and the gauge.

The Peirce decomposition is The Peirce Decomposition of $M_4(\mathbb{C})$; the rank-two split is The Rank-Two Split of $M_4(\mathbb{C})$; the root and the gauge are The Frame, the Simple Nilpotents and the Gauge of $M_4(\mathbb{C})$; the adjoint action and the flag variety are The Adjoint Action, the Orbits and the Flag Variety of $M_4(\mathbb{C})$; the topology is The Topology of $M_4(\mathbb{C})$.

Rank-One Idempotents

Definition. An idempotent is a matrix $P$ with

$$ P^2=P. $$

Its rank is the dimension of its image. A rank-one idempotent has the form

$$ P=u\,\varphi,\qquad \varphi(u)=1, $$

where $u\in\mathbb{C}^4$ is a nonzero vector and $\varphi\in(\mathbb{C}^4)^*$ is a linear form with $\varphi(u)=1$. The image is the line $L=\mathbb{C}u$, and the kernel is the hyperplane $H=\ker\varphi$, which does not contain $u$.

The affine bundle. The map

$$ \pi:\{P:\operatorname{rank}P=1,\ P^2=P\}\longrightarrow\mathbb{P}^3, \qquad P\longmapsto\operatorname{im}P, $$

sends a rank-one idempotent to its image line. The fibre over a line $L$ is

$$ \pi^{-1}(L)=\{\varphi\in L^*:\varphi(u)=1\}, $$

an affine space of dimension $3$: the linear forms on $L$ with value $1$ on $u$. The total space is an affine bundle of rank $3$ over $\mathbb{P}^3$, of complex dimension

$$ 3+3=6. $$

The chart. In the frame where $u=e_1$, a rank-one idempotent with image the first coordinate line is

$$ P=E_{11}+aE_{12}+bE_{13}+cE_{14} =\begin{pmatrix} 1&a&b&c\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0 \end{pmatrix}, \qquad a,b,c\in\mathbb{C}. $$

Indeed $P^2=P$, the image is $\mathbb{C}e_1$, and the kernel is

$$ x_1+ax_2+bx_3+cx_4=0. $$

The three coordinates $(a,b,c)$ are the fibre coordinates over the line $[e_1]$. The transition functions between charts are affine, because the condition $\varphi(u)=1$ is affine in the frame.

The fibre is not a vector space. The fibre over $L$ is an affine space, not a vector space: there is no canonical origin. The orthogonal projection onto $L$,

$$ P_L=u(u^*u)^{-1}u^*, $$

is a distinguished point of the fibre, the point where the kernel is $L^\perp$. It gives a global section of the affine bundle and identifies the fibre with the vector space $L^\perp$-valued forms; the section depends on the Hermitian metric.

The orthogonal idempotents. The idempotents with $P^*=P$ are the orthogonal projections. A rank-one orthogonal projection is determined by its image line, so the orthogonal rank-one idempotents form the Grassmannian

$$ \operatorname{Gr}(1,4)=\mathbb{P}^3, $$

of dimension $3$. They are the zero section of the affine bundle; the general rank-one idempotents have a nonzero complement coordinate.

Rank-$p$ Idempotents

The homogeneous space. Let $P$ be an idempotent of rank $p$. The image $U=\operatorname{im}P$ and the kernel $W=\ker P$ are complementary:

$$ \mathbb{C}^4=U\oplus W,\qquad \dim U=p,\quad \dim W=4-p. $$

Conversely, every direct sum decomposition determines a unique idempotent, the projection onto $U$ along $W$. The group $GL_4$ acts transitively on the rank-$p$ idempotents by conjugation, and the stabiliser of $P$ is the block-diagonal group

$$ GL_p\times GL_{4-p} =\left\{\begin{pmatrix}A&0\\0&D\end{pmatrix}:A\in GL_p,\ D\in GL_{4-p}\right\}. $$

In the projective group the stabiliser is its image

$$ P(GL_p\times GL_{4-p})=(GL_p\times GL_{4-p})/\mathbb{C}^\times I. $$

Hence the rank-$p$ idempotents form the homogeneous space

$$ \mathcal{I}_p=PGL_4/P(GL_p\times GL_{4-p}), $$

of complex dimension

$$ \dim\mathcal{I}_p=15-\bigl(p^2+(4-p)^2-1\bigr) =16-p^2-(4-p)^2 =2p(4-p). $$

The values are

$$ p=0:\ 0,\qquad p=1:\ 6,\qquad p=2:\ 8,\qquad p=3:\ 6,\qquad p=4:\ 0. $$

The bundle over the Grassmannian. The projection

$$ \pi:\mathcal{I}_p\longrightarrow\operatorname{Gr}(p,4),\qquad P\longmapsto\operatorname{im}P, $$

is a fibre bundle. The fibre over a $p$-plane $U$ is the affine space of complements $W$ to $U$, equivalently the affine space of linear maps

$$ B:W\longrightarrow U, $$

the graphs of which are the complements. The fibre has dimension

$$ p(4-p), $$

and the total dimension is

$$ p(4-p)+p(4-p)=2p(4-p). $$

The fibre is contractible, so the total space $\mathcal{I}_p$ is homotopy equivalent to $\operatorname{Gr}(p,4)$.

The chart. In a decomposition $\mathbb{C}^4=U\oplus W$ with $U$ spanned by the first $p$ coordinates, every idempotent with image $U$ has the block form

$$ P=\begin{pmatrix}I_p&B\\0&0\end{pmatrix}, \qquad B\in\operatorname{Hom}(W,U)\cong\mathbb{C}^{p\times(4-p)}. $$

The square is $P$ because the lower-left block is zero and the upper-left block is the identity; the kernel is the graph of $-B$ in $W$. The matrix $B$ is the fibre coordinate; the orthogonal projection is $B=0$.

The orthogonal idempotents. The self-adjoint idempotents of rank $p$ are the orthogonal projections, determined by their image $p$-plane. They form the Grassmannian

$$ \operatorname{Gr}(p,4),\qquad \dim=p(4-p), $$

the zero section of the affine bundle. The general idempotents have the complement coordinate $B\ne0$; the affine bundle is the total space, and the Grassmannian is the compact skeleton.

The rank-one case. For $p=1$,

$$ \mathcal{I}_1\longrightarrow\mathbb{P}^3 $$

is the affine bundle of rank $3$ of the previous section; for $p=3$ it is the dual affine bundle over $\mathbb{P}^3$; for $p=2$ it is the affine bundle of rank $4$ over $\operatorname{Gr}(2,4)$, of dimension $8$.

The Tangent Space

The tangent space at $P$. Differentiate the equation $P(t)^2=P(t)$ at $t=0$:

$$ X P+P X=X, $$

where $X=\dot P(0)$. The tangent space of the idempotent variety at $P$ is therefore

$$ T_P\mathcal{I}_p=\{X\in M_4:XP+PX=X\}. $$

The Peirce form. In the decomposition $\mathbb{C}^4=U\oplus W$, write

$$ P=\begin{pmatrix}I_p&B\\0&0\end{pmatrix}, \qquad X=\begin{pmatrix}X_{11}&X_{12}\\X_{21}&X_{22}\end{pmatrix}, \qquad B:W\to U . $$

The condition $XP+PX=X$ is equivalent to

$$ X_{11}=-BX_{21},\qquad X_{22}=X_{21}B, $$

with $X_{12}:W\to U$ and $X_{21}:U\to W$ free. Hence

$$ \dim T_P\mathcal{I}_p=p(4-p)+p(4-p)=2p(4-p). $$

Equivalently, in the Peirce decomposition of $P$, the tangent space is $PMQ\oplus QMP$, the two off-diagonal corners. The diagonal Peirce corners $PMP\oplus QMQ$ are the normal directions for the trace form. At the orthogonal projection $B=0$ the coordinate blocks and the Peirce corners coincide, and the tangent space is the off-diagonal block space.

Smoothness. Each rank-$p$ stratum is a homogeneous space $PGL_4/P$, so it is smooth, and the tangent space has the expected dimension. The rank is locally constant on the idempotent variety, because $\operatorname{rank}P=\operatorname{tr}P$ is continuous and integer-valued; the strata are therefore open and closed, and there is no closure order between different ranks.

The Rank Stratification

The strata. The idempotents split by rank:

$$ \mathcal{I}=\bigsqcup_{p=0}^{4}\mathcal{I}_p, \qquad \mathcal{I}_p=\{P:P^2=P,\ \operatorname{rank}P=p\}. $$

The strata are the orbits of $PGL_4$; the rank is the only invariant, because every rank-$p$ idempotent is conjugate to $\operatorname{diag}(I_p,0)$.

The strata are clopen. The rank of an idempotent is its trace, $\operatorname{rank}P=\operatorname{tr}P$, which is continuous and integer-valued on the idempotent variety. Hence each $\mathcal{I}_p$ is both open and closed, and the closure of $\mathcal{I}_p$ is $\mathcal{I}_p$ itself:

$$ \overline{\mathcal{I}_p}=\mathcal{I}_p. $$

The variety is the disjoint union of the five connected components $\mathcal{I}_0,\dots,\mathcal{I}_4$. The zero idempotent and the identity are isolated points; the rank-one and rank-three strata are six-dimensional; the rank-two stratum is eight-dimensional. There is no closure order between different ranks: a limit of rank-$p$ idempotents has rank $p$.

The rank-two case. The rank-two idempotents have dimension $8$; the orthogonal ones form $\operatorname{Gr}(2,4)$ of dimension $4$. In the standard frame the two chiral idempotents $P_+$ and $P_-$ of The Two Chiralities of $M_4(\mathbb{C})$ are two particular orthogonal rank-two idempotents, points of the Grassmannian, not components. For a general root they are non-Hermitian projections and correspond to transverse pairs of planes, not to points of $\operatorname{Gr}(2,4)$. The general rank-two idempotent has a nonzero mass block; the Peirce table is The Rank-Two Split of $M_4(\mathbb{C})$.

The decomposition. The rank-$p$ idempotent gives the Peirce corners

$$ p^2+p(4-p)+(4-p)p+(4-p)^2=16, $$

with dimensions

$$ p^2,\quad p(4-p),\quad (4-p)p,\quad (4-p)^2. $$

For $p=1$: $1+3+3+9$; for $p=2$: $4+4+4+4$; for $p=3$: $9+3+3+1$. The geometry of the idempotent is the geometry of its Peirce corners.

The Flag Variety and the Partial Flags

The full flag variety. A full flag is a chain

$$ 0=F_0\subset F_1\subset F_2\subset F_3\subset F_4=\mathbb{C}^4, \qquad \dim F_k=k. $$

The flag variety is

$$ \mathcal{F}=GL_4/B, $$

where $B$ is the Borel subgroup of upper triangular matrices. Its dimension is

$$ \dim\mathcal{F}=16-10=6. $$

The flag variety is smooth, projective and simply connected.

The Grassmannians. The Grassmannian $\operatorname{Gr}(p,4)$ is the space of $p$-planes; its dimension is

$$ \dim\operatorname{Gr}(p,4)=p(4-p). $$

In particular,

$$ \operatorname{Gr}(1,4)=\mathbb{P}^3,\quad \operatorname{Gr}(2,4),\quad \dim4,\quad \operatorname{Gr}(3,4)=\mathbb{P}^3 . $$

The idempotents are transverse pairs, not flags. A rank-$p$ idempotent is a pair $(U,W)$ with

$$ U\oplus W=\mathbb{C}^4,\qquad \dim U=p, $$

a transverse pair. A partial flag of type $(p,4-p)$ is a nested pair

$$ U\subset W,\qquad \dim U=p,\quad \dim W=4-p, $$

which is a different object. The idempotent variety is the open subset of $\operatorname{Gr}(p,4)\times\operatorname{Gr}(4-p,4)$ where $U\cap W=0$, of dimension $2p(4-p)$; the partial flag variety is the closed incidence subset where $U\subset W$, of dimension $p(4-p)+(4-2p)p$ for $p\le2$. The two are different: the idempotent variety is affine and noncompact, while the partial flag variety is projective and compact. The product of Grassmannians is a compactification of the idempotent variety; the partial flag variety is a different closed subvariety of the product.

The incidence relations. The partial flag variety of type $(p,q)$ with $p

$$ \mathcal{F}_{p,q}=\{(U,V):U\subset V,\ \dim U=p,\ \dim V=q\}, $$

of dimension

$$ p(4-p)+(q-p)(4-q). $$

The full flag maps to it with fibres the product of the full flag varieties of the subquotients $U$, $V/U$ and $\mathbb{C}^4/V$. The standard incidence correspondences are the projections to the two Grassmannians; the fibres are Grassmannians of the subquotients. These are the nested-flag analogues of the transverse-pair bundle of the idempotents.

The gauge as the fibre of the frame bundle. The projection

$$ GL_4\longrightarrow\mathbb{P}^3,\qquad S\longmapsto S e_1, $$

is a principal bundle with fibre the parabolic stabiliser of the line. The maximal torus $T\cong(\mathbb{C}^\times)^3$ of $PGL_4$ is the gauge torus; the frame bundle over the root orbit,

$$ PGL_4\longrightarrow PGL_4/T, $$

has fibre $T$, and the gauge is that fibre. The count is

$$ \underbrace{15}_{\text{frame group}} = \underbrace{12}_{\text{root orbit}} + \underbrace{3}_{\text{gauge}}. $$

The idempotents $E_{ii}$ of a frame are the coordinate projections; the gauge rescales the frame and acts on the off-diagonal nilpotents, while the diagonal idempotents are invariant.

The Biquaternion Case

At $n=2$ the idempotents are simpler. The rank-one idempotents of $M_2(\mathbb{C})$ form

$$ PGL_2/P(GL_1\times GL_1), $$

of dimension

$$ 2\cdot1\cdot1=2, $$

an affine bundle of rank $1$ over $\mathbb{P}^1$. In a chart,

$$ P=\begin{pmatrix}1&a\\0&0\end{pmatrix}, \qquad a\in\mathbb{C}, $$

the image is the first coordinate line, and the kernel is the line $x_1+ax_2=0$. The orthogonal rank-one idempotents form $\mathbb{P}^1$.

The rank-two case. The identity is the only rank-two idempotent; the zero matrix is the only rank-zero idempotent. The rank-one idempotents are the two-dimensional family; the coarse Peirce corners are $1+1+1+1$.

The comparison. The full algebra has the affine bundles of ranks $3,4,3$ over $\mathbb{P}^3,\operatorname{Gr}(2,4),\mathbb{P}^3$; the biquaternion algebra has the affine bundle of rank $1$ over $\mathbb{P}^1$. The pattern is the same: the idempotents of rank $p$ form the affine bundle of complements over $\operatorname{Gr}(p,n)$, of dimension $2p(n-p)$.

Summary

The idempotents of rank $p$ in $M_4(\mathbb{C})$ form the homogeneous space

$$ \mathcal{I}_p=PGL_4/P(GL_p\times GL_{4-p}), \qquad \dim_\mathbb{C}\mathcal{I}_p=2p(4-p). $$

For $p=1$ this is an affine bundle of rank $3$ over $\mathbb{P}^3$, of dimension $6$; for $p=2$ it is an affine bundle of rank $4$ over $\operatorname{Gr}(2,4)$, of dimension $8$; for $p=3$ it is the dual affine bundle over $\mathbb{P}^3$, of dimension $6$. The fibre over a $p$-plane is the affine space of complements, of dimension $p(4-p)$, contractible; the orthogonal projections form the zero section $\operatorname{Gr}(p,4)$, and $\mathcal{I}_p$ is homotopy equivalent to $\operatorname{Gr}(p,4)$. The tangent space at $P$ is the sum of the two off-diagonal Peirce corners $\{X:XP+PX=X\}$, of dimension $2p(4-p)$; the rank strata are clopen components, with $\overline{\mathcal{I}_p}=\mathcal{I}_p$ because $\operatorname{rank}P=\operatorname{tr}P$ is locally constant. The full flag variety $GL_4/B$ has dimension $6$ and is simply connected; the idempotents are transverse pairs of a $p$-plane and a $(4-p)$-plane, not nested flags. The gauge torus $T\cong(\mathbb{C}^\times)^3$ is the fibre of the frame bundle over the root orbit, and $15=12+3$.

$$ \boxed{\ \dim_\mathbb{C}\mathcal{I}_p=2p(4-p),\quad \mathcal{I}_1\to\mathbb{P}^3\ \text{affine rank }3,\quad \mathcal{I}_p\simeq\operatorname{Gr}(p,4),\quad \dim_\mathbb{C}\mathcal{F}=6. } $$

Object Complex dimension
$\mathcal{I}_0$ $0$
$\mathcal{I}_1$ $6$
$\mathcal{I}_2$ $8$
$\mathcal{I}_3$ $6$
$\mathcal{I}_4$ $0$
$\operatorname{Gr}(p,4)$ $p(4-p)$
Full flag variety $6$
Gauge torus $3$

Summary of Notation

Symbol Meaning
$\mathcal{I}_p$ the idempotents of rank $p$
$P(GL_p\times GL_{4-p})$ the projective parabolic stabiliser, with Levi $GL_p\times GL_{4-p}$
$\operatorname{Gr}(p,4)$ the Grassmannian of $p$-planes
$\mathcal{F}=GL_4/B$ the full flag variety
$T$ the gauge torus
$B$ the complement coordinate of an idempotent

Further Reading