The Frame, the Simple Nilpotents and the Gauge of $M_n(\mathbb{C})$
Introduction
A root fixes the flag, and a frame fixes one vector in each line of the flag; the freedom in that choice is a torus, one parameter per simple nilpotent. The root therefore gives the idempotents and the lines, and the gauge gives the vectors and the units. The two together are the frame.
The article constructs the frame from a basis adapted to the spectral flag, states the matrix-unit rule, isolates the simple nilpotents $E_{k,k+1}$, computes the gauge action and its cocycle, and counts the parameters: $n(n-1)$ complex from the root and $n-1$ complex from the gauge. At $n=2$ the count is four real and two real, the six real dimensions of the biquaternion frame. The article also identifies the older "second root" with a point of the gauge, which is the precise correction of the biquaternion presentation.
The Adapted Frame
A frame adapted to the flag is a basis with one vector in each spectral line. Let $H^{n}=I$ have simple spectrum, with projectors $P_k$ and lines $V_k=\operatorname{im}P_k$. Choose $v_k\in V_k$, $v_k\ne0$, for $k=1,\dots,n$. Then $(v_1,\dots,v_n)$ is a basis of $\mathbb{C}^{n}$, and
$$ P_kv_m=\delta_{km}v_k,\qquad Hv_k=\omega_kv_k. $$
Let $\ell_k$ be the dual basis of the dual space, $\ell_k(v_m)=\delta_{km}$, and put
$$ E_{kl}=v_k\ell_l. $$
Proposition. The $n^{2}$ elements $E_{kl}$ are a basis of $M_n(\mathbb{C})$, they satisfy
$$ E_{kl}E_{mn}=\delta_{lm}E_{kn},\qquad E_{kk}=P_k,\qquad H=\sum_{k=1}^{n}\omega_kE_{kk}, $$
and $\operatorname{tr}E_{kl}=\delta_{kl}$.
Proof. $E_{kl}E_{mn}=v_k\ell_l(v_m)\ell_n=\delta_{lm}v_k\ell_n=\delta_{lm}E_{kn}$. On $v_m$, $E_{kk}$ acts as $\delta_{km}v_k$, so $E_{kk}=P_k$. The sum $\sum_k\omega_kE_{kk}$ acts on $v_m$ as $\omega_mv_m=Hv_m$. The trace is $\ell_l(v_k)=\delta_{kl}$.
The frame is the choice of the $v_k$. The root fixes the lines; the frame fixes the vectors. Different frames of the same flag give different matrices $E_{kl}$ and the same idempotents $E_{kk}=P_k$.
The Simple Nilpotents
The superdiagonal units are nilpotent of index two, and every invertible element is a product of consecutive ones. Put
$$ N_k=E_{k,k+1},\qquad k=1,\dots,n-1. $$
Then
$$ N_k^{2}=E_{k,k+1}E_{k,k+1}=0, $$
and for $k $$
E_{kl}=N_kN_{k+1}\cdots N_{l-1}.
$$ The nilpotents $N_k$ are the simple nilpotents of the frame. There are $n-1$ of them, one for each gap between consecutive lines of the flag. The simple nilpotents generate the frame. The algebra generated by $N_1,\dots,N_{n-1}$ and their transposes $N_k^{\mathsf T}$ is the whole of $M_n(\mathbb{C})$: the products of consecutive $N$'s give all $E_{kl}$ with $k The chain is the Peirce chain. The nilpotent $N_k$ maps the line $V_{k+1}$ to the line $V_k$ and kills the other lines. In the Peirce notation of The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$, $N_k$ is a nonzero element of $A_{k,k+1}$. The chain $N_1,\dots,N_{n-1}$ is the sequence of maps that steps down the flag. A rescaling of the frame vectors rescales the matrix units by the ratio of the two scales. Replace $v_k$ by $t_kv_k$ with $t_k\in\mathbb{C}^{\times}$. The dual basis is replaced by $t_k^{-1}\ell_k$, so $$
E_{kl}=v_k\ell_l\longmapsto t_kt_l^{-1}E_{kl}.
$$ The idempotents are fixed, $$
E_{kk}\longmapsto E_{kk},
$$ and so is the root $H=\sum_k\omega_kE_{kk}$. The group acting is the torus $$
T=\{t=(t_1,\dots,t_n)\in(\mathbb{C}^{\times})^{n}\},
$$ and the diagonal $t_1=\cdots=t_n$ acts trivially, so the effective group is $$
T/\mathbb{C}^{\times}\cong(\mathbb{C}^{\times})^{n-1}.
$$ This is the gauge group of the frame. One parameter per simple nilpotent. Put $$
\lambda_k=t_kt_{k+1}^{-1},\qquad k=1,\dots,n-1.
$$ Then $$
N_k\longmapsto\lambda_kN_k,
$$ and every other unit is fixed by the $\lambda$'s through the cocycle $$
E_{kl}\longmapsto\lambda_k\lambda_{k+1}\cdots\lambda_{l-1}\,E_{kl},\qquad k and the inverse product for $k>l$. The composite $E_{kl}$ has no independent scale: its scale is the product of the consecutive ones. This is the cocycle of the gauge, and it is the reason the gauge has $n-1$ parameters rather than one per corner. The cocycle forbids an extra parameter. Suppose that in addition to the $t_k$ one tries to rescale a single composite $E_{kl}$, $k $$
E_{kl}=E_{k,k+1}\cdots E_{l-1,l},
$$ so the scale of $E_{kl}$ is the product of the scales of the factors, and $\mu$ is determined by them. Equivalently, the gauge action on the invertibles is the coboundary of the $0$-cochain $t$, and a coboundary has no freedom beyond the cochain modulo constants. The idempotents do not depend on the gauge. The parameters $\lambda_k$ act on the nilpotents and on the off-diagonal units, never on the diagonal. In particular the projectors $P_k$ and the spectral decomposition $H=\sum_k\omega_kP_k$ are gauge invariant. A statement that depends on $\lambda$ is a statement about a frame, not about the algebra or the idempotents. A frame modulo overall scale has $n^{2}-1$ complex dimensions. The set of bases of $\mathbb{C}^{n}$ is $GL_n(\mathbb{C})$, of complex dimension $n^{2}$; two bases that differ by an overall scalar give the same matrix units, so the frame is a point of $PGL_n(\mathbb{C})$, of complex dimension $n^{2}-1$. The root and the gauge split this count. The map from frames to roots, $(v_k)\mapsto H=\sum_k\omega_kE_{kk}$, has fibre the torus $T/\mathbb{C}^{\times}\cong(\mathbb{C}^{\times})^{n-1}$: the frames that give the same root are exactly the rescalings of the $v_k$. Hence $$
\underbrace{n^{2}-1}_{\text{frame}/\,\mathbb{C}^{\times}}
=
\underbrace{n(n-1)}_{\text{root}}
+
\underbrace{n-1}_{\text{gauge}} .
$$ The case $n=2$ gives the six real dimensions. For $n=2$, $$
n^{2}-1=3,\qquad n(n-1)=2,\qquad n-1=1,
$$ all complex. In real dimensions the frame modulo scale has $6$, the root has $4$ and the gauge has $2$: the four real parameters of the biquaternion root and the two real parameters of the complex number $\lambda$. This is the six-dimensional basis of the biquaternion frame. The general case. For $n=3,4,5,6$ the complex counts are The root column is the dimension of $\mathcal{R}_n$ in Roots of Unity and the Lagrange Idempotents of $M_n(\mathbb{C})$; the gauge column is $n-1$, one per simple nilpotent. At $n=2$ the frame is the companion pair of the biquaternion algebra. With $H=\operatorname{diag}(1,-1)$ in the frame $(v_1,v_2)$, $$
N_1=E_{12},\qquad N_1^{\mathsf T}=E_{21} .
$$ These are the two companion nilpotents of The Companion Nilpotents of the Biquaternion Algebra, written in the frame. They satisfy $N_1^{2}=(N_1^{\mathsf T})^{2}=0$, $N_1N_1^{\mathsf T}=P_1$ and $N_1^{\mathsf T}N_1=P_2$. The last two identities are the biquaternion relations $\vec\Upsilon_{+-}\vec\Upsilon_{-+}=\tilde\Pi_+$ and $\vec\Upsilon_{-+}\vec\Upsilon_{+-}=\tilde\Pi_-$, and the general rule behind them is $E_{kl}E_{lk}=E_{kk}$. The basis is the choice of one of the two elements. Choosing $N_1$ fixes $N_1^{\mathsf T}$ by the normalisation, which is the bijection of The Peirce Basis of a Peirce Decomposition of the Biquaternion Algebra, §Existence: The Nilpotent Route. The gauge is the reciprocal rescaling $N_1\mapsto\lambda N_1$, $N_1^{\mathsf T}\mapsto\lambda^{-1}N_1^{\mathsf T}$, and it is the whole freedom. An involution anticommuting with $H$ is one point of that gauge. Let $R$ satisfy $$
R^{2}=I,\qquad RH=-HR .
$$ The anticommutation forces $R$ to be off-diagonal, $R=aE_{12}+bE_{21}$, and $R^{2}=abI=I$ forces $b=a^{-1}$. Hence $$
R=aE_{12}+a^{-1}E_{21},\qquad a\in\mathbb{C}^{\times}.
$$ Under the gauge $E_{12}\mapsto\lambda E_{12}$, $E_{21}\mapsto\lambda^{-1}E_{21}$, the element $R$ is carried to $a\lambda E_{12}+a^{-1}\lambda^{-1}E_{21}$, which is the same family with $a$ replaced by $a\lambda$. The orbit of $R$ is the whole gauge fibre, and the parameter $a$ is gauge, not a root. What $R$ carries and what it does not. It names one nonzero element in each mixed Peirce line, $R$ having nonzero entries exactly in $A_{12}$ and $A_{21}$, and it normalises the two reciprocally. It supplies no idempotent: those are $P_\pm=(I\pm H)/2$, produced by $H$ alone. So $R$ is a shorthand for a normalised pair $(N_1,N_1^{\mathsf T})$, and the pair is the basis. The biquaternion articles state the construction in the pair, as the choice of a nonzero element of the mixed line; they introduce no element beyond the root and the basis. For $n>2$ there is no single such element. The gauge has $n-1$ complex parameters, and the analogue of $R$ is the full torus of rescalings of the $v_k$. An element anticommuting with $H$ can be built at $n=4$, but its orbit has dimension two and does not carry the third parameter. No single element plays the role that $R$ plays at $n=2$, because the flag has more than two lines. The Fourier frame. Take $v_k=(1,\omega^{k-1},\omega^{2(k-1)})^{\mathsf T}$ with $\omega=e^{2\pi i/3}$, the eigenvectors of the cyclic shift of Roots of Unity and the Lagrange Idempotents of $M_n(\mathbb{C})$. Then $$
E_{kl}=\frac{1}{3}v_kv_l^{*}
$$ because the $v_k$ are orthogonal with $\|v_k\|^{2}=3$, and the dual functional is $\ell_l=\tfrac13v_l^{*}$. The three simple nilpotents are $$
N_1=\frac13v_1v_2^{*},\qquad N_2=\frac13v_2v_3^{*},
$$ and $$
N_1N_2=\frac19v_1\,(v_2^{*}v_2)\,v_3^{*}=\frac13v_1v_3^{*}=E_{13},
$$ which is the cocycle at work: the scale of $E_{13}$ is the product of the scales of $N_1$ and $N_2$. A gauge transformation. Rescale $v_1\mapsto t_1v_1$, $v_2\mapsto t_2v_2$ and $v_3\mapsto t_3v_3$. Then $$
N_1\longmapsto\frac{t_1}{t_2}N_1=\lambda_1N_1,\qquad
N_2\longmapsto\frac{t_2}{t_3}N_2=\lambda_2N_2,
$$ and $$
E_{13}\longmapsto\frac{t_1}{t_3}E_{13}=\lambda_1\lambda_2E_{13}.
$$ The two parameters $\lambda_1,\lambda_2$ are independent; the third scale $\lambda_1\lambda_2$ is not. An adapted frame $(v_1,\dots,v_n)$ gives the matrix units $E_{kl}=v_k\ell_l$, with $E_{kl}E_{mn}=\delta_{lm}E_{kn}$, $E_{kk}=P_k$ and $H=\sum_k\omega_kE_{kk}$. The simple nilpotents $N_k=E_{k,k+1}$, $n-1$ of them, square to zero and generate the frame; every composite is their product. The gauge $v_k\mapsto t_kv_k$ acts by $E_{kl}\mapsto(t_k/t_l)E_{kl}$, an effective torus $(\mathbb{C}^{\times})^{n-1}$ with one parameter per simple nilpotent, subject to the cocycle that fixes the composites. The frame modulo scale has $n^{2}-1$ complex dimensions, split as $n(n-1)$ for the root and $n-1$ for the gauge; at $n=2$ this is $6=4+2$ real, the biquaternion count. An involution of the frame anticommuting with $H$ is a point of this gauge, not a second primitive.The Gauge Action
The Count of the Frame
$n$
frame $/\,\mathbb{C}^{\times}$
root
gauge
$3$
$8$
$6$
$2$
$4$
$15$
$12$
$3$
$5$
$24$
$20$
$4$
$6$
$35$
$30$
$5$
The Companion Nilpotents at $n=2$
An Explicit Frame for $n=3$
Summary
Summary of Notation
Symbol
Meaning
$v_k$
a nonzero vector in the spectral line $V_k$
$\ell_k$
the dual functional, $\ell_k(v_m)=\delta_{km}$
$E_{kl}=v_k\ell_l$
the matrix unit of the frame
$N_k=E_{k,k+1}$
the simple nilpotent, $N_k^{2}=0$
$t_k$
the gauge scale of $v_k$
$\lambda_k=t_kt_{k+1}^{-1}$
the gauge parameter on $N_k$
$T$
the torus of rescalings, effective $(\mathbb{C}^{\times})^{n-1}$
Further Reading