The Geometry of the Idempotents of $M_6(\mathbb{C})$
Introduction
An idempotent of $M_6(\mathbb{C})$ is a matrix $E$ with
$$ E^2=E. $$
It is a projector: $V=\operatorname{im}E\oplus\ker E$, and $E$ is the projection onto its image along its kernel. The rank-$p$ idempotents form a smooth variety
$$ \mathcal E_p=\{E:E^2=E,\ \operatorname{rank}E=p\}, $$
of complex dimension
$$ \dim\mathcal E_p=2p(6-p). $$
For $p=1$ the variety is an affine bundle of rank $5$ over $\mathbb{P}^5$, of dimension $10$; for general $p$ it is the homogeneous space
$$ \mathcal E_p\cong PGL_6/P(GL_p\times GL_{6-p}), $$
a fibre bundle over $\operatorname{Gr}(p,6)$ with the contractible fibre of complements. The full flag variety $GL_6/B$ has dimension $15$; the partial flags and their incidence relations organise the idempotents, and the gauge is the fibre of the frame bundle.
The article assumes the Grassmannians and the orbit–stabilizer count of The Adjoint Action, the Orbits and the Flag Variety of $M_6(\mathbb{C})$ and the Peirce decomposition of The Peirce Decomposition of $M_6(\mathbb{C})$. It uses the elementary geometry of homogeneous spaces. The idempotents of the biquaternion algebra are The Square Roots of One and the Idempotents of the Biquaternion Algebra; the spectral geometry is The Sextic and the Spectral Strata of $M_6(\mathbb{C})$.
The Idempotents
Definition and Rank
Definition. An idempotent is $E\in M_6(\mathbb{C})$ with $E^2=E$. Its rank is the dimension of its image.
Theorem. If $E^2=E$, then
$$ V=\operatorname{im}E\oplus\ker E, \qquad E|_{\operatorname{im}E}=I,\quad E|_{\ker E}=0, $$
and the rank is the trace. Conversely, every direct sum decomposition $V=U\oplus W$ defines the idempotent $E$ with image $U$ and kernel $W$.
Proof. For $v\in V$, $v=Ev+(v-Ev)$; $Ev\in\operatorname{im}E$ and $v-Ev\in\ker E$ because $E(v-Ev)=Ev-E^2v=0$. The intersection is zero: if $Ev=0$ and $Ev=v$, then $v=0$. On the image, $E(Ev)=E^2v=Ev$; on the kernel, $E=0$. The trace is the sum of the eigenvalues, which are $1$ on the image and $0$ on the kernel.
Corollary (the eigenvalues). The eigenvalues of an idempotent are $0$ and $1$; the rank is the multiplicity of $1$.
Proof. The minimal polynomial divides $t^2-t=t(t-1)$.
The Rank-One Idempotents
Theorem. A rank-one idempotent is
$$ E=u\varphi, \qquad \varphi(u)=1, $$
with $u\in\mathbb{C}^6$ and $\varphi\in(\mathbb{C}^6)^*$. Its image is the line $\mathbb{C}u$, and its kernel is the hyperplane $\ker\varphi$.
Proof. A rank-one matrix is $u\varphi$; $E^2=u\varphi(u)\varphi=\varphi(u)E$, so $E^2=E$ iff $\varphi(u)=1$. The image is $\mathbb{C}u$; the kernel is $\ker\varphi$.
Theorem (the affine bundle). The rank-one idempotents form an affine bundle of rank $5$ over $\mathbb{P}^5$:
$$ \pi:\mathcal E_1\longrightarrow\mathbb{P}^5, \qquad E\longmapsto\operatorname{im}E . $$
The fibre over the line $L$ is the affine space of complements $H$ to $L$ in $\mathbb{C}^6$, of dimension $5$. Hence
$$ \dim\mathcal E_1=5+5=10 . $$
Proof. The image of a rank-one idempotent is a line, so the map is well defined and surjective. The fibre over $L$ is the set of kernels $H$ complementary to $L$, equivalently the set of splittings $V=L\oplus H$. The set of complements to $L$ is an affine space modelled on $\operatorname{Hom}(V/L,L)$, of dimension $5$. The total dimension is the sum.
Corollary (the affine structure). Choose a line $L$ and a complement $H_0$. Every other complement is the graph of a linear map $L\to H_0$; the affine coordinates are the $5$ parameters of that map.
Proof. A complement $H$ transverse to $L$ is the graph of a unique linear map $L\to H_0$; the graph is a $5$-dimensional affine space.
Corollary (the explicit form). In a basis with $L=\mathbb{C}e_1$ and $H_0=\langle e_2,\dots,e_6\rangle$, the rank-one idempotents with image $L$ are
$$ E=\begin{pmatrix}1&a_2&\cdots&a_6\\0&0&\cdots&0\\\vdots&\vdots&&\vdots\\0&0&\cdots&0\end{pmatrix}, \qquad (a_2,\dots,a_6)\in\mathbb{C}^5 . $$
Proof. $E=e_1\varphi$ with $\varphi(e_1)=1$ and $\varphi(e_j)=a_j$; the matrix is the row $(1,a_2,\dots,a_6)$ in the first row.
The Rank-$p$ Idempotents
Theorem. The rank-$p$ idempotents form the homogeneous space
$$ \mathcal E_p\cong PGL_6/P(GL_p\times GL_{6-p}), $$
of complex dimension
$$ \dim\mathcal E_p=2p(6-p). $$
Proof. $PGL_6$ acts transitively on rank-$p$ idempotents: every such idempotent is conjugate to $E_p=\operatorname{diag}(I_p,0)$. The stabilizer of $E_p$ is the image in $PGL_6$ of the block-diagonal $GL_p\times GL_{6-p}$, of dimension $p^2+(6-p)^2-1$. The orbit dimension is
$$ 35-\bigl(p^2+(6-p)^2-1\bigr)=36-p^2-(6-p)^2=2p(6-p). $$
Corollary (the table).
| $p$ | $2p(6-p)$ | Description |
|---|---|---|
| $0$ | $0$ | the zero idempotent |
| $1$ | $10$ | affine bundle over $\mathbb{P}^5$ |
| $2$ | $16$ | affine bundle over $\operatorname{Gr}(2,6)$ |
| $3$ | $18$ | affine bundle over $\operatorname{Gr}(3,6)$ |
| $4$ | $16$ | affine bundle over $\operatorname{Gr}(4,6)$ |
| $5$ | $10$ | affine bundle over $\operatorname{Gr}(5,6)=\mathbb{P}^5$ |
| $6$ | $0$ | the identity |
Proof. Substitute $p$ in $2p(6-p)$.
Theorem (the bundle over the Grassmannian). The image map
$$ \pi_p:\mathcal E_p\longrightarrow\operatorname{Gr}(p,6), \qquad E\longmapsto\operatorname{im}E $$
is a fibre bundle with fibre the affine space of complements, of dimension $p(6-p)$. Hence
$$ \dim\mathcal E_p=p(6-p)+p(6-p)=2p(6-p). $$
Proof. The image is a $p$-plane; the kernel is a complementary $(6-p)$-plane. The set of complements to a fixed $p$-plane is an affine space modelled on $\operatorname{Hom}(\mathbb{C}^6/V_p,V_p)$, of dimension $p(6-p)$. The total dimension is the sum.
Corollary (the contractible fibre). The fibre is an affine space, hence contractible; the variety $\mathcal E_p$ is homotopy equivalent to the Grassmannian $\operatorname{Gr}(p,6)$.
Proof. A fibre bundle with contractible fibre has the homotopy type of the base.
Remark (the two descriptions). The homogeneous-space description $PGL_6/P(GL_p\times GL_{6-p})$ and the affine-bundle description over $\operatorname{Gr}(p,6)$ are the same variety: the base is the image Grassmannian, the fibre is the complement, and the structure group is the affine general linear group of the complement.
The Flag Variety
The Full Flag Variety
Definition. A complete flag is a chain
$$ 0\subset V_1\subset V_2\subset\cdots\subset V_5\subset\mathbb{C}^6, \qquad \dim V_i=i . $$
The full flag variety is
$$ \mathcal F=GL_6/B, $$
where $B$ is the Borel subgroup of upper triangular matrices.
Theorem. $\mathcal F$ is a smooth projective variety of dimension
$$ \dim\mathcal F=36-21=15 . $$
Proof. $GL_6$ acts transitively on complete flags; the stabilizer of the standard flag is $B$, of dimension $21$ (six diagonal and fifteen upper off-diagonal parameters); the quotient has dimension $15$.
Corollary (the iterated bundle). The flag variety is an iterated projective bundle with fibres
$$ \mathbb{P}^5,\ \mathbb{P}^4,\ \mathbb{P}^3,\ \mathbb{P}^2,\ \mathbb{P}^1, $$
and
$$ 15=5+4+3+2+1 . $$
Proof. Choose $V_1$ in $\mathbb{P}^5$; then $V_2/V_1$ in $\mathbb{P}^4$; then $V_3/V_2$ in $\mathbb{P}^3$; then $V_4/V_3$ in $\mathbb{P}^2$; then $V_5/V_4$ in $\mathbb{P}^1$.
The Partial Flags
Definition. A partial flag of type $(p_1,\dots,p_k)$ is a chain
$$ 0\subset V_{p_1}\subset\cdots\subset V_{p_k}\subset\mathbb{C}^6, \qquad \dim V_{p_i}=p_i . $$
Theorem. The partial flag variety of type $(p)$ is the Grassmannian
$$ \mathcal F_{(p)}=\operatorname{Gr}(p,6), \qquad \dim=p(6-p). $$
For the type $(p,q)$ with $p $$
\dim\mathcal F_{(p,q)}=p(6-p)+(q-p)(6-q).
$$ Proof. Choose $V_p$ (dimension $p(6-p)$); then $V_q/V_p$ is a $(q-p)$-plane in $\mathbb{C}^6/V_p\cong\mathbb{C}^{6-p}$, of dimension $(q-p)(6-p-(q-p))=(q-p)(6-q)$. Corollary (the table). Proof. Apply the formula. Theorem. The full flag variety fibres over every partial flag variety: $$
\pi_{p_1,\dots,p_k}:\mathcal F\longrightarrow\mathcal F_{(p_1,\dots,p_k)},
$$ with fibre the product of the flag varieties of the successive quotients. Proof. A complete flag projects to its subspaces of the prescribed dimensions; the fibre consists of the flags refining the partial flag. Corollary (the rank-$p$ idempotent and the flag). A rank-$p$ idempotent determines the two-step partial flag $$
0\subset\operatorname{im}E\subset\mathbb{C}^6,
$$ but not a complete flag; the complete flag requires the further splitting of the image and the kernel. The fibre of $\mathcal F\to\operatorname{Gr}(p,6)$ over the image is the product of the full flag varieties of the $p$-plane and the $(6-p)$-quotient, of dimension $$
\frac{p(p-1)}2+\frac{(6-p)(5-p)}2 .
$$ Proof. The image and the quotient are the two blocks of the idempotent; a complete flag refining $0\subset\operatorname{im}E\subset V$ is a flag of the image times a flag of the quotient. Remark (idempotent versus flag). An idempotent of rank $p$ is more than its image: it also chooses the kernel, a complement. The Grassmannian records the image; the idempotent records the splitting; the flag records the refinement. Definition. A frame of $\mathbb{C}^6$ is an ordered basis $(v_1,\dots,v_6)$. The frame bundle over the flag variety is $$
\pi:GL_6\longrightarrow GL_6/B=\mathcal F,
$$ sending a frame to the flag of its coordinate spans. Theorem. $\pi$ is a principal $B$-bundle: the fibre over a flag is the Borel subgroup $B$ of dimension $21$. Proof. Two frames give the same flag iff they differ by an upper triangular change of basis, that is by an element of $B$. Corollary (the gauge). The gauge is the maximal torus of $B$ modulo the center, $$
(\mathbb{C}^\times)^5,
$$ of dimension $5$. It acts on the frame by rescaling the basis vectors. Proof. The diagonal torus of $GL_6$ is $(\mathbb{C}^\times)^6$; modulo the center $\mathbb{C}^\times I_6$ it is $(\mathbb{C}^\times)^5$. The rescaling of the frame preserves the flag. Theorem. Over the root $H$, the eigenframe bundle is a principal $(\mathbb{C}^\times)^6$-bundle; modulo the overall scale it is a principal $(\mathbb{C}^\times)^5$-bundle, and the gauge is its fibre. Proof. The eigenbases of $H$ are the frames in which $H$ is diagonal; two such frames differ by independent rescalings, the torus $(\mathbb{C}^\times)^6$. The overall scale acts trivially on the projective frame, so the gauge is $(\mathbb{C}^\times)^6/\mathbb{C}^\times\cong(\mathbb{C}^\times)^5$. Corollary (the dimensions). $$
\dim GL_6=36=\dim\mathcal F+\dim B=15+21,
\qquad
\dim B=21=15+6 .
$$ The Borel has dimension $21$; its maximal torus has dimension $6$; the unipotent radical has dimension $15$; the gauge is the torus modulo the center, of dimension $5$. Proof. The Borel is the semidirect product of the torus (dimension $6$) and the unipotent radical (dimension $15$); the flag variety has dimension $15$. Remark (the three fibres). The frame bundle over the flag variety has fibre the Borel ($21$); the eigenframe bundle over the root has fibre the torus ($6$, or $5$ projectively); the idempotent bundle over the Grassmannian has fibre the affine space of complements ($p(6-p)$). The gauge is the torus fibre; the flag fibre is the Borel; the idempotent fibre is the complement. The rank-$p$ idempotents form the smooth variety $$
\mathcal E_p\cong PGL_6/P(GL_p\times GL_{6-p}),
\qquad
\dim_{\mathbb{C}}\mathcal E_p=2p(6-p),
$$ an affine bundle over the Grassmannian $\operatorname{Gr}(p,6)$ with fibre the contractible affine space of complements, of dimension $p(6-p)$. $$
\boxed{\ \dim\mathcal E_1=10=5+5,\qquad \dim\mathcal E_p=2p(6-p),\qquad \dim GL_6/B=15 .\ }
$$ For $p=1$ the variety is an affine bundle of rank $5$ over $\mathbb{P}^5$; for $p=3$ it has dimension $18$; for $p=6$ it is the single identity. The full flag variety $GL_6/B$ has dimension $15$, with the iterated bundle $5+4+3+2+1$. The partial flags are the Grassmannians and their incidences; an idempotent determines the two-step flag of its image, and the complete flag refines it. The gauge $(\mathbb{C}^\times)^5$ is the fibre of the frame bundle over the root; the Borel is the fibre over the flag.
Type
Dimension
$(1)$
$5$
$(2)$
$8$
$(3)$
$9$
$(1,2)$
$5+4=9$
$(2,4)$
$8+2\cdot2=12$
$(1,3)$
$5+2\cdot3=11$
$(1,2,3)$
$5+4+3=12$
$(1,2,3,4,5)$
$15$
The Incidence
The Gauge and the Frame Bundle
The Frame Bundle
The Gauge as the Fibre
Summary
Object
Dimension
$\mathcal E_1$
$10$
$\mathcal E_2$
$16$
$\mathcal E_3$
$18$
$\mathcal E_p$
$2p(6-p)$
$\operatorname{Gr}(p,6)$
$p(6-p)$
Full flag variety
$15$
Borel $B$
$21$
Gauge $(\mathbb{C}^\times)^5$
$5$
Summary of Notation
Symbol
Meaning
$\mathcal E_p$
the rank-$p$ idempotents
$P(GL_p\times GL_{6-p})$
the projective Levi stabilizer
$\operatorname{Gr}(p,6)$
the Grassmannian of $p$-planes
$\mathcal F=GL_6/B$
the full flag variety
$\mathcal F_{(p_1,\dots,p_k)}$
the partial flag variety
$B$
the Borel subgroup of upper triangular matrices
Further Reading