The Frame, the Simple Nilpotents and the Real Gauge of $M_n(\mathbb{R})$
Introduction
A complete real frame carries a family of simple nilpotents $N=\sum_{k}c_{k}E_{k,k+1}$ with all $c_{k}\ne0$, and over $\mathbb{R}$ the coefficients live in $(n-1)$ copies of $\mathbb{R}^{\times}$, which is a space with $2^{n-1}$ components.
The article defines the frame, defines the simple nilpotents relative to it, proves that such an $N$ is nilpotent of index $n$ and rank $n-1$, computes the real gauge group of the frames, $\mathrm{GL}_{n}(\mathbb{R})$ with stabiliser the diagonal torus of dimension $n$, shows that the simple nilpotents over a fixed frame form a torus of dimension $n-1$ with $2^{n-1}$ components, shows that the gauge fixing $N$ is the one-dimensional group of scalars, and proves that every regular real nilpotent is conjugate to the shift $E_{12}+\cdots+E_{n-1,n}$. It assumes The Peirce Basis and the Corners over $\mathbb{R}$ for the frame and its matrix units, and it complements The Frame, the Simple Nilpotents and the Gauge of $M_n(\mathbb{C})$, where the same counts are complex.
The Frame
A frame is a complete family of orthogonal idempotents summing to the identity.
$$ P_{k}P_{l}=\delta_{kl}P_{k},\qquad \sum_{k=1}^{n}P_{k}=\mathrm{I},\qquad \operatorname{rk}P_{k}=1 . $$
Each $P_{k}$ projects onto a line $V_{k}$, the lines are independent, and $\mathbb{R}^{n}=V_{1}\oplus\cdots\oplus V_{n}$. The standard frame is $F_{k}=E_{kk}$, with $V_{k}$ the $k$-th coordinate line.
A frame is the same datum as a real basis, taken up to scale. A choice of nonzero $v_{k}\in V_{k}$ gives a real basis, and a different choice replaces $v_{k}$ by $t_{k}v_{k}$ with $t_{k}\ne0$; the frame is unchanged. So a frame is a basis modulo the diagonal scaling group $(\mathbb{R}^{\times})^{n}$.
The frame carries a system of matrix units. With $\ell_{k}$ the dual form on the lines, $E_{kl}=v_{k}\ell_{l}$ obeys $E_{ij}E_{kl}=\delta_{jk}E_{il}$ and $\sum_{k}E_{kk}=\mathrm{I}$, and $E_{kk}=P_{k}$. In this basis the elements $E_{kl}$ are the elementary arrays.
The Simple Nilpotents
A simple nilpotent relative to the frame is an element whose only nonzero Peirce components lie on the first superdiagonal, all of them nonzero.
$$ N=\sum_{k=1}^{n-1}c_{k}E_{k,k+1},\qquad c_{k}\in\mathbb{R}^{\times}. $$
In the Peirce coordinates of the frame this says $N=\sum_{k}N_{k,k+1}$ with $N_{k,k+1}\in A_{k,k+1}$ and $N_{k,k+1}\ne0$; in the adapted basis $N$ is the array with $c_{k}$ in position $(k,k+1)$ and zeros elsewhere.
Proposition. Such an $N$ satisfies
$$ N^{j}=\sum_{k=1}^{n-j}\Big(\prod_{i=0}^{j-1}c_{k+i}\Big)E_{k,k+j}\quad (1\le j\le n-1),\qquad N^{n}=0, $$
so $N$ is nilpotent of index $n$, of rank $n-1$, and $\operatorname{rk}N^{j}=n-j$.
Proof. The power $N^{j}$ selects chains of $j$ consecutive superdiagonal steps, hence the entries on the $j$-th superdiagonal with the products of the $c$'s. Every product is nonzero, so $N^{j}\ne0$ for $j\le n-1$, and no chain of $n$ steps exists, so $N^{n}=0$. The array $N^{j}$ has $n-j$ nonzero entries, one per row, hence rank $n-j$.
A simple nilpotent has the characteristic polynomial $x^{n}$. All its eigenvalues are $0$, its trace is $0$, and its rank is $n-1$, so it is a single Jordan block of size $n$ over $\mathbb{R}$.
The chain of images is the frame reversed. The powers give
$$ \operatorname{im}N^{j}=\operatorname{span}(v_{1},\dots,v_{n-j}),\qquad \ker N^{j}=\operatorname{span}(v_{1},\dots,v_{j}), $$
so $N$ moves each line of the frame one step, and the flag it defines is the frame read backwards.
The Relations with the Idempotents
The simple nilpotent is characterised by its Peirce relations with the frame. The corner rule gives
$$ E_{kk}N=NE_{k+1,k+1}\quad (k=1,\dots,n-1),\qquad E_{kk}N=0\ \text{for}\ k=n,\qquad NE_{11}=0 , $$
and, conversely, an element satisfying these relations and lying in the sum of the corners $A_{k,k+1}$ is of the form above.
Proof. Multiply $N=\sum_{j}c_{j}E_{j,j+1}$ on the left by $E_{kk}$ and on the right by $E_{k+1,k+1}$; the table $E_{kk}E_{j,j+1}=\delta_{kj}E_{j,j+1}$ and $E_{j,j+1}E_{k+1,k+1}=\delta_{j,k}E_{j,j+1}$ leaves $c_{k}E_{k,k+1}$ on both sides.
The relations say that $N$ shifts the frame and kills the ends. $N$ maps $V_{k+1}$ to $V_{k}$ for $k\le n-1$, and it kills $V_{1}$; so $N$ is a chain map of the frame into itself, decreasing the index by one.
The coefficients are the gauge-dependent data. A different choice of the vectors $v_{k}$ changes the units by $E_{kl}\mapsto t_{k}t_{l}^{-1}E_{kl}$ and hence
$$ c_{k}\longmapsto \frac{t_{k}}{t_{k+1}}c_{k}, $$
so the individual numbers $c_{k}$ are not attached to the frame; only their orbit under the diagonal group is.
The Real Gauge of the Frames
Conjugation by $\mathrm{GL}_{n}(\mathbb{R})$ acts transitively on the frames. For two frames there is $S\in\mathrm{GL}_{n}(\mathbb{R})$ with $P'_{k}=SP_{k}S^{-1}$ for all $k$, built from two real bases adapted to the frames.
The stabiliser of the standard frame is the diagonal subgroup.
$$ D=\{\operatorname{diag}(t_{1},\dots,t_{n}):t_{k}\in\mathbb{R}^{\times}\}\cong(\mathbb{R}^{\times})^{n}, $$
of dimension $n$. The quotient is the variety of frames,
$$ \text{frames}=\mathrm{GL}_{n}(\mathbb{R})/D,\qquad \dim=n^{2}-n . $$
The group is real and not complex. The conjugating matrix is real because both adapted bases are real; a complex matrix carrying the standard frame to a real frame is real up to a diagonal factor, and the diagonal factor acts trivially.
The counts at $n=3$ and $n=4$. At $n=3$ the frame variety has dimension $9-3=6$; at $n=4$ it has dimension $16-4=12$. The diagonal torus has dimension $n$ in both cases, and the two numbers add to $n^{2}$, the dimension of the group.
The Torus of Simple Nilpotents
Over a fixed frame the simple nilpotents form $(n-1)$ copies of $\mathbb{R}^{\times}$.
$$ \mathcal{N}(P)=\Big\{\sum_{k=1}^{n-1}c_{k}E_{k,k+1}:c_{k}\in\mathbb{R}^{\times}\Big\}\cong(\mathbb{R}^{\times})^{n-1}, $$
of real dimension $n-1$.
The space has $2^{n-1}$ connected components. Each factor $\mathbb{R}^{\times}$ has two components, the positive and the negative reals, so a connected component is labelled by the signs of the $n-1$ coefficients:
$$ \varepsilon=(\operatorname{sgn}c_{1},\dots,\operatorname{sgn}c_{n-1})\in\{\pm1\}^{n-1}. $$
Two simple nilpotents lie in the same component exactly when all their coefficients have the same signs.
The diagonal gauge acts transitively on the whole torus. The group $D$ acts by $c_{k}\mapsto t_{k}t_{k+1}^{-1}c_{k}$, and the map $D\to(\mathbb{R}^{\times})^{n-1}$, $t\mapsto(t_{1}t_{2}^{-1},\dots,t_{n-1}t_{n}^{-1})$, is surjective: given the ratios, choose $t_{1}$ and then $t_{k+1}=t_{k}(t_{k}t_{k+1}^{-1})^{-1}$. So all $2^{n-1}$ sign classes lie in one orbit of the frame-preserving gauge, and the components are a topological fact about $\mathbb{R}^{\times}$, not a dynamical one.
The count is the orbit–stabiliser count inside $D$. The stabiliser is computed in the next section and has dimension $1$, so
$$ \dim D=n=\underbrace{(n-1)}_{\text{simple nilpotents}}+\underbrace{1}_{\text{stabiliser}}. $$
The Gauge that Fixes the Nilpotent
The gauge transformations fixing a simple nilpotent are the scalar matrices.
$$ \{S\in D:SN=NS\}=\{t\,\mathrm{I}:t\in\mathbb{R}^{\times}\}, $$
a group of dimension $1$, with $2$ components.
Proof. Write $S=\operatorname{diag}(t_{1},\dots,t_{n})$ and compare the $(k,k+1)$ entries of $SN$ and $NS$: they are $t_{k}c_{k}$ and $t_{k+1}c_{k}$, so equality for all $k$ gives $t_{k}=t_{k+1}$, since $c_{k}\ne0$. Hence all $t_{k}$ are equal.
The scalars act trivially on the frame and on the Peirce coordinates. A scalar matrix commutes with every element of the algebra, so it fixes the frame, the units and the nilpotent; the one-dimensional group is the kernel of the action and not a genuine symmetry of the data.
A negative scalar acts trivially on the nilpotent and on the units. It fixes every $E_{kl}$, since $t_{k}t_{l}^{-1}=1$, and changes the vectors $v_{k}$ by a common sign, which is invisible in the frame.
The Canonical Form
The shift $N_{0}=E_{12}+E_{23}+\cdots+E_{n-1,n}$ is the simple nilpotent with all $c_{k}=1$.
$$ N_{0}=\begin{pmatrix}0&1& & \\ &0&1& \\ & & \ddots&1\\ & & &0\end{pmatrix}. $$
Theorem. Every regular nilpotent of $M_{n}(\mathbb{R})$, that is, every real matrix with characteristic polynomial $x^{n}$ and rank $n-1$, is conjugate over $\mathbb{R}$ to $N_{0}$.
Proof. Let $N$ be such a matrix. Such a vector exists: if $N^{n-1}=0$ then every Jordan block has size at most $n-1$, so there are at least two blocks and the rank is at most $n-2$. The vectors $v,Nv,\dots,N^{n-1}v$ are independent: a relation of least degree would give $N^{m}v\in\operatorname{span}(v,\dots,N^{m-1}v)$, hence $N^{n-1}v=0$. In the reversed basis $N^{n-1}v,\dots,Nv,v$ the matrix of $N$ is the shift, so with $S$ the matrix of that basis,
$$ S^{-1}NS=N_{0}. $$
The basis is real because $N$ and $v$ are real, so the conjugating matrix is in $\mathrm{GL}_{n}(\mathbb{R})$.
The canonical form is the statement that the frame's nilpotent has no moduli over $\mathbb{R}$. Every regular nilpotent has a real Jordan form with one block, and one block of size $n$ is the shift. The simple nilpotents of a fixed frame are the shifts of that frame, and the gauge carries any one of them to any other, so the coefficients $c_{k}$ are pure gauge.
The theorem does not hold with a fixed frame. A regular nilpotent carries its own frame, the flag $\operatorname{im}N^{n-1}\subset\cdots\subset\operatorname{im}N\subset\mathbb{R}^{n}$ refined by lines, and that frame is what the conjugating matrix must respect; the freedom is the $n$-dimensional diagonal torus, and the $n-1$ coefficients are its coordinates.
The Checks at $n=3$ and $n=4$
The characteristic polynomial of a random regular nilpotent is $x^{n}$. Building $N=SN_{0}S^{-1}$ with $S$ a random real matrix with integer entries in a small range, the characteristic polynomial computed with numpy is $x^{3}$ at $n=3$ and $x^{4}$ at $n=4$, and the rank is $n-1$ in every trial. Building $N$ directly as $\sum_{k}c_{k}E_{k,k+1}$ with random nonzero $c_{k}$ gives the same answers.
The powers behave as predicted. At $n=4$ with all $c_{k}=1$ the ranks of $N^{0},N^{1},N^{2},N^{3},N^{4}$ are $4,3,2,1,0$, which is the sequence $n-j$; and $N^{4}=0$ while $N^{3}\ne0$.
The dimension count closes at every $n$. The frame variety has dimension $n^{2}-n$, the nilpotent torus over a frame has dimension $n-1$, and the diagonal stabiliser has dimension $n$; the identity $n^{2}-n+(n-1)+1=n^{2}$ is the dimension of $\mathrm{GL}_{n}(\mathbb{R})$.
Summary
A frame of $M_{n}(\mathbb{R})$ is a complete family of orthogonal rank-one idempotents summing to $\mathrm{I}$, equivalently a real basis taken up to diagonal scaling. Relative to a frame with matrix units $E_{kl}$, a simple nilpotent is $N=\sum_{k=1}^{n-1}c_{k}E_{k,k+1}$ with all $c_{k}\ne0$; it obeys $E_{kk}N=NE_{k+1,k+1}$, has powers with entries on the successive superdiagonals, is nilpotent of index $n$, and has rank $n-1$, hence characteristic polynomial $x^{n}$. The real gauge is conjugation by $\mathrm{GL}_{n}(\mathbb{R})$, which acts transitively on the frames with stabiliser the diagonal torus $D\cong(\mathbb{R}^{\times})^{n}$ of dimension $n$, so the frame variety has dimension $n^{2}-n$: $6$ at $n=3$ and $12$ at $n=4$. Over a fixed frame the simple nilpotents form $(\mathbb{R}^{\times})^{n-1}$, of dimension $n-1$ and with $2^{n-1}$ components labelled by the signs of the coefficients, and the diagonal gauge acts transitively on that torus. The gauge fixing a simple nilpotent is the group of scalar matrices, of dimension $1$ and with $2$ components, so $n=(n-1)+1$ inside $D$. The canonical form is the shift $E_{12}+\cdots+E_{n-1,n}$, and every real regular nilpotent is conjugate to it, with a real conjugating matrix built from the Krylov basis $v,Nv,\dots,N^{n-1}v$; this is confirmed at $n=3$ and $n=4$ by the characteristic polynomial $x^{n}$ and the rank $n-1$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $P_{k}$, $V_{k}$ | The idempotents of the frame and their lines |
| $E_{kl}$ | The matrix unit $v_{k}\ell_{l}$ of the frame |
| $N=\sum_{k}c_{k}E_{k,k+1}$ | A simple nilpotent, with $c_{k}\in\mathbb{R}^{\times}$ |
| $D\cong(\mathbb{R}^{\times})^{n}$ | The diagonal stabiliser of the standard frame |
| $\mathcal{N}(P)\cong(\mathbb{R}^{\times})^{n-1}$ | The simple nilpotents over the frame $P$ |
| $\mathrm{GL}_{n}(\mathbb{R})/D$ | The frame variety, of dimension $n^{2}-n$ |
| $N_{0}=E_{12}+\cdots+E_{n-1,n}$ | The shift, canonical form of a regular nilpotent |
| $v,Nv,\dots,N^{n-1}v$ | The Krylov basis of a regular nilpotent |
Further Reading
- Richard S. Pierce, Associative Algebras (Springer, 1982), for frames, matrix units and the conjugation that carries one frame to another.
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge, 2013), for nilpotent matrices, Jordan forms over $\mathbb{R}$ and the Krylov basis.
- James E. Humphreys, Linear Algebraic Groups (Springer, 1975), for the diagonal torus, homogeneous spaces and the orbit–stabiliser count.
- The Peirce Basis and the Corners over $\mathbb{R}$, for the corners and the frame that the simple nilpotents use.
- The Frame, the Simple Nilpotents and the Gauge of $M_n(\mathbb{C})$, for the complex version of the same construction.
- The Matrix Algebra $M_n(\mathbb{R})$, for the matrix units and the transpose over $\mathbb{R}$.