The Triangular Decomposition $h\oplus n_+\oplus n_-$

Introduction

The diagonal, the strictly upper triangular part and the strictly lower triangular part cut $M_n(\mathbb{R})$ into three pieces of dimensions $n$, $n(n-1)/2$ and $n(n-1)/2$, and the sum is direct with total $n^{2}$.

The article defines the standard flag, proves that its stabiliser is the upper triangular algebra, counts the three pieces, proves that the upper triangular algebra is a subalgebra and that the strictly upper triangular part is a nilpotent ideal of index $n$, measures the brackets $[n_+,n_+]$ and $[n_+,n_-]$, and reads the decomposition as the infinitesimal Iwasawa picture. It assumes The Fine Decomposition against the Diagonal for the lines $L_{ij}$ and their weights, and The Two Halves: Symmetric and Antisymmetric for the antisymmetric part that appears in the Iwasawa sum.

The Standard Flag

Definition. Let $e_1,\dots,e_n$ be the standard basis of $\mathbb{R}^{n}$ and put $V_k=\operatorname{span}\{e_1,\dots,e_k\}$. The chain

$$ 0=V_0\subset V_1\subset V_2\subset\dots\subset V_n=\mathbb{R}^{n} $$

is the standard flag, and $\dim V_k=k$.

A complete flag is a chain of one subspace of each dimension. The chain above has $n$ steps and no room for a further subspace: if $V_{k-1}\subset W\subset V_k$ then $W$ is one of the two ends, because the dimensions differ by one.

The stabiliser is smaller over $\mathbb{R}$ than over $\mathbb{C}$. Over $\mathbb{C}$ the same chain has the complex upper triangular group as stabiliser, of complex dimension $n(n+1)/2$ and real dimension $n(n+1)$. Over $\mathbb{R}$ the stabiliser has real dimension $n(n+1)/2$, which is also the dimension of the symmetric half.

The Stabiliser Is the Upper Triangular Algebra

Definition. $UT_n=\{A\in M_n(\mathbb{R}):A_{ij}=0\ \text{for}\ i>j\}$.

Theorem. $A$ stabilises the standard flag, that is $AV_k\subseteq V_k$ for every $k$, if and only if $A\in UT_n$.

Proof. The condition $AV_k\subseteq V_k$ says that the image of $e_j$ lies in $V_j$ for every $j$, hence that the column $j$ of $A$ has no entry below row $j$. That is the condition $A_{ij}=0$ for $i>j$. Conversely, if $A$ is upper triangular then $Ae_j\in V_j$, and a linear combination of $e_1,\dots,e_k$ maps into $V_k$.

The stabiliser is closed under the product and under inversion. If $A$ and $B$ stabilise the flag, so does $AB$, because $ABV_k\subseteq AV_k\subseteq V_k$. If $A$ stabilises the flag and is invertible, so does $A^{-1}$, because $A$ induces an isomorphism $V_k\to V_k$ at each step.

The algebra splits as the diagonal plus the strictly upper part.

$$ UT_n=h\oplus n_+,\qquad \dim UT_n=n+\frac{n(n-1)}{2}=\frac{n(n+1)}{2}. $$

The measured dimensions are $3$ at $n=2$, $6$ at $n=3$ and $10$ at $n=4$, equal to $n(n+1)/2$ in each case.

The Three Pieces

Definition. $n_+$ is the set of strictly upper triangular matrices and $n_-$ the set of strictly lower triangular matrices:

$$ n_+=\operatorname{span}\{E_{ij}:ij\}. $$

Each strictly triangular part has dimension $n(n-1)/2$. The pairs $ij$ number the same. So

$$ \dim n_+=\dim n_-=\frac{n(n-1)}{2}. $$

The three pieces are the graded pieces of the fine grading grouped by the sign of the weight. A line $L_{ij}$ lies in $n_+$ when $\varepsilon_i-\varepsilon_j$ is positive, in $n_-$ when it is negative, and the diagonal is the zero weight. This is the grouping announced in The Fine Decomposition against the Diagonal.

Theorem. $M_n(\mathbb{R})=h\oplus n_+\oplus n_-$ and the sum is direct.

Proof. Every matrix is the sum of its diagonal, its strictly upper part and its strictly lower part, and each of the three is read off from a disjoint set of entries. So the three subspaces span and their pairwise intersections are zero.

The dimension count closes.

$$ n+\frac{n(n-1)}{2}+\frac{n(n-1)}{2}=n+n(n-1)=n^{2}. $$

Measured as the rank of the union of the three bases, the sum has dimension $4$ at $n=2$, $9$ at $n=3$ and $16$ at $n=4$.

The Upper Triangular Algebra

Proposition. $UT_n$ is a subalgebra of $M_n(\mathbb{R})$, and $n_+$ is a two-sided ideal of it.

Proof. The product of two upper triangular matrices is upper triangular: $(AB)_{ij}=\sum_{k}A_{ik}B_{kj}$, and a term is nonzero only when $i\le k\le j$, which forces $i\le j$. If in addition $A\in n_+$ then $i

The ideal $n_+$ is also an ideal for the commutator. For $H\in h$ and $X\in n_+$ the bracket $[H,X]$ is strictly upper triangular, and for $U\in UT_n$ and $X\in n_+$ the bracket $[U,X]=UX-XU$ lies in $n_+$ because both products do.

The quotient $UT_n/n_+$ is the diagonal. $h$ maps isomorphically onto the quotient, so the diagonal is the reductive part of the triangular algebra and $n_+$ is its nilpotent radical.

Nilpotence and the Index

Theorem. $n_+^{n}=0$ and $n_+^{n-1}\ne0$.

Proof. A product of $m$ strictly upper triangular matrices is a sum of products $E_{i_0i_1}E_{i_1i_2}\cdots E_{i_{m-1}i_m}$ with $i_0

The index is exactly $n$, and the measurement agrees. On the sum $N$ of all strictly upper matrix units, the powers had norm $1.7$ and then $0$ at $n=3$, and norms $2.4$, $2.4$, $1.0$ and then $0$ at $n=4$. The first vanishing power is the $n$-th in each case.

Nilpotence is what makes the exponential a polynomial. For $N\in n_+$ the series $\exp N=\sum_{m

The Bracket of $n_+$ with Itself

Proposition. $[n_+,n_+]\subseteq n_+$, and the inclusion is strict for $n\ge3$.

Proof. Both products of two strictly upper triangular matrices are strictly upper triangular, so the bracket is. For strictness, count: $[E_{ij},E_{kl}]$ is nonzero only when $j=k$ or $l=i$, and for $i

Measured, the span of the brackets is short of the whole. The brackets of the standard basis of $n_+$ span dimension $1$ out of $3$ at $n=3$, and dimension $3$ out of $6$ at $n=4$. At $n=2$ the span is $0$ out of $1$.

The deficiency has a one-line cause. A Lie algebra generated by one element is spanned by that element alone, since $[x,x]=0$, so it is abelian. For $n\ge3$ the algebra $n_+$ is not abelian, and its derived algebra is strictly smaller than itself.

Iterating the bracket walks away from the diagonal. The $m$-th term of the lower central series of $n_+$ is spanned by the units $E_{ij}$ with $j-i\ge m$, so each step removes one superdiagonal. The unit $E_{1n}$ survives to the last step, and the series reaches zero at $m=n$, which is the index.

The Bracket of $n_+$ with $n_-$

The bracket of the two triangular parts is not confined to the diagonal. With $il$,

$$ [E_{ij},E_{kl}]=\delta_{jk}E_{il}-\delta_{li}E_{kj}, $$

and the right-hand side lies in $h$ only when both indices meet, namely when $j=k$ and $i=l$, which is the pair $[E_{ij},E_{ji}]=E_{ii}-E_{jj}$.

What the bracket spans has a diagonal part of dimension $n-1$. The diagonal components are the differences $E_{ii}-E_{jj}$, and they span the trace-zero diagonal, of dimension $n-1$. Measured, that diagonal part has dimension $1$ at $n=2$, $2$ at $n=3$ and $3$ at $n=4$.

The whole bracket is larger than its diagonal part. Measured as the span of all brackets $[E_{ij},E_{kl}]$ with $il$, the dimension is $1$ at $n=2$, $6$ at $n=3$ and $13$ at $n=4$. The remaining components lie in $n_+$ and in $n_-$: at $n=3$ the span is the diagonal of trace zero together with $E_{12}$, $E_{21}$, $E_{23}$, $E_{32}$.

So the triangular decomposition is not a symmetric pair. A symmetric pair needs $[n_+,n_-]\subseteq h$, and here the bracket leaves $h$ as soon as $n\ge3$. The pair that does satisfy the three inclusions is $(UT_n,n_+)$, and the pair $(\mathfrak{gl}_n(\mathbb{R}),\mathfrak{so}(n))$ of Symmetric Pairs and Cartan Pairs: the Reductivity Criterion over $\mathbb{R}$ is another.

The correct statement about $[n_+,n_-]$ is therefore an inclusion into the whole algebra, with a diagonal shadow of dimension $n-1$. The shadow is what the trace detects: the trace of a bracket is zero, so every diagonal component has trace zero.

Abelian Only at $n\le2$

Proposition. $n_+$ is abelian if and only if $n\le2$.

Proof. $n_+$ has dimension $n(n-1)/2$, which is $0$ for $n=1$ and $1$ for $n=2$; a space of dimension at most one is abelian. For $n\ge3$ the units $E_{12}$ and $E_{23}$ are in $n_+$ and $[E_{12},E_{23}]=E_{13}\ne0$, so $n_+$ is not abelian.

The same holds for $n_-$, by transposition. The transpose exchanges $n_+$ and $n_-$ and changes the sign of the bracket, so the two are abelian at the same orders.

The measurement at $n=2$ gives the value zero. The bracket span has dimension $0$ at $n=2$ and dimension $1$ at $n=3$, which is the statement in numbers.

The first non-abelian case is $n=3$, and it is the smallest nilpotent Heisenberg-like example. There $n_+$ has basis $E_{12}$, $E_{23}$, $E_{13}$ with $[E_{12},E_{23}]=E_{13}$ and $E_{13}$ central, which is the three-dimensional Heisenberg algebra in upper triangular form.

The Infinitesimal Bruhat and Iwasawa Picture

The quotient $\mathfrak{gl}_n(\mathbb{R})/(h\oplus n_+)$ is $n_-$. The subspace $h\oplus n_+$ is the Lie algebra of the stabiliser of the flag, of dimension $n(n+1)/2$, so the quotient has dimension $n^{2}-n(n+1)/2=n(n-1)/2$, which is the dimension of $n_-$. The map $n_-\to\mathfrak{gl}_n/(h\oplus n_+)$ is an isomorphism of vector spaces, and it identifies $n_-$ with the tangent space of the flag variety $\mathrm{GL}_n(\mathbb{R})/UT_n$ at the base point.

This is the infinitesimal Bruhat statement. The flag variety has dimension $n(n-1)/2$, and the tangent space at the base point is the strictly lower triangular part, whose entries are the $n(n-1)/2$ coordinates of a first-order deformation of the flag.

The Iwasawa decomposition of the Lie algebra uses the other triangular part. With $\mathfrak{k}=\mathfrak{so}(n)=\operatorname{Skew}_n$, $\mathfrak{a}=h$ and $\mathfrak{n}=n_+$,

$$ \mathfrak{gl}_n(\mathbb{R})=\mathfrak{k}\oplus\mathfrak{a}\oplus\mathfrak{n},\qquad \frac{n(n-1)}{2}+n+\frac{n(n-1)}{2}=n^{2}, $$

and the measured dimension of the sum is $4$, $9$, $16$ at $n=2,3,4$. The three pieces are the compact part, the split part and the nilpotent part, and the antisymmetric half of The Two Halves: Symmetric and Antisymmetric is the compact one.

The two triangular decompositions are mirror images. $h\oplus n_+$ is the stabiliser of the flag and $h\oplus n_-$ the stabiliser of the opposite flag, built from $e_n,e_{n-1},\dots,e_1$. The intersection of the two is $h$, and the two nilpotent parts are exchanged by the transpose. So the algebra has two dual triangular readings of the same grading, and the diagonal is what they share.

Summary

The standard flag has stabiliser $UT_n=h\oplus n_+$, of dimension $n+n(n-1)/2=n(n+1)/2$. The three pieces $h$, $n_+$ and $n_-$ have dimensions $n$, $n(n-1)/2$ and $n(n-1)/2$, and their direct sum is the whole algebra of dimension $n^{2}$, measured as $4$, $9$, $16$ at $n=2,3,4$. $UT_n$ is a subalgebra and $n_+$ is a nilpotent ideal of index $n$, with $n_+^{n}=0$ and $n_+^{n-1}\ne0$. The bracket $[n_+,n_+]$ is strictly smaller than $n_+$ for $n\ge3$, spanning $1$ of $3$ dimensions at $n=3$. The bracket $[n_+,n_-]$ is not contained in $h$: its diagonal part is the trace-zero diagonal of dimension $n-1$, and its whole span has dimension $1$, $6$, $13$ at $n=2,3,4$. The two triangular parts are abelian only at $n\le2$. The Iwasawa sum $\mathfrak{so}(n)\oplus h\oplus n_+$ is the algebra, and the quotient by $h\oplus n_+$ is the tangent space of the flag variety.

Summary of Notation

Symbol Meaning
$V_k$ the span of $e_1,\dots,e_k$, the steps of the standard flag
$h$ the diagonal subalgebra, of dimension $n$
$n_+$ the strictly upper triangular part, of dimension $n(n-1)/2$
$n_-$ the strictly lower triangular part, of dimension $n(n-1)/2$
$UT_n=h\oplus n_+$ the upper triangular algebra, the stabiliser of the flag
$N=\sum_k E_{k,k+1}$ the shift, with $N^{n-1}=E_{1n}$ and $N^{n}=0$
$[X,Y]=XY-YX$ the commutator, under which $n_+$ is a nilpotent ideal

Further Reading