2
$\begingroup$

I wish to have a proof for the following result:

Let $U_n$ be an $n\times n$ upper shift matrix, and $L_n = U_n^T$ be a lower shift matrix. For example, $$ U_5 = \begin{pmatrix} 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 & 0 \end{pmatrix}, \quad L_5 = \begin{pmatrix} 0 & 0 & 0 & 0 & 0 \\ 1 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 \end{pmatrix}. $$ Then the Lie algebra generated by $U_n$ and $L_n$ over $\mathbb{C}$ is isomorphic to $\mathfrak{so}(n, \mathbb{C})$ when $n$ is odd, and is isomorphic to $\mathfrak{sp}(n, \mathbb{C})$ when $n$ is even.

I tested this numerically with GAP over the field $\mathbb{Q}$: the Lie algebra generated by $U_{2n+1}$ and $L_{2n+1}$ over $\mathbb{Q}$ is of type $B_n$, while the Lie algebra generated by $U_{2n}$ and $L_{2n}$ over $\mathbb{Q}$ is of type $C_n$.

$\endgroup$
10
  • 1
    $\begingroup$ For which bilinear forms? For the standard quadratic form $\mathfrak{so}(n,\mathbb{C})$ is the space of skew-symmetric matrices, which does not contain $L_n$ or $U_n$. $\endgroup$
    – abx
    Sep 1, 2021 at 4:29
  • 1
    $\begingroup$ @abx That should be part of the question. It should be some weird bilinear form which I could not determine. $\endgroup$ Sep 1, 2021 at 4:31
  • 1
    $\begingroup$ Oh, I see. Maybe you could say that $L_n$ and $U_n$ generate a Lie algebra isomorphic to $\mathfrak{so}(n,\mathbb{C})$ or $\mathfrak{sp}(n,\mathbb{C})$. $\endgroup$
    – abx
    Sep 1, 2021 at 5:13
  • $\begingroup$ @abx Thanks for the suggestion! I've made the change in the text. $\endgroup$ Sep 1, 2021 at 5:16
  • 1
    $\begingroup$ Here's a roadmap to a solution. (a) compute the space $B_n$ of invariant quadratic (for $n$ odd) or alternating forms by $U_n$ and $L_n$. Then $B_n$ should be 1-dimensional and generated by a non-degenerate form (otherwise the conclusion fails), say $q$. So, the Lie subalgebra $g_n$ generated by $\{U_n,L_n\}$ preserves $q$, i.e., is contained in so(q), resp sp(q). (b) Show the latter inclusion is an equality, e.g., computing enough iterated brackets to obtain a good lower bound on the dimension of $g_n$? $\endgroup$
    – YCor
    Sep 1, 2021 at 6:36

1 Answer 1

4
$\begingroup$

Let ${\mathfrak g} $ be the Lie algebra generated by $U_n$ and $L_n$. It's easy to check that $U_n$ and $L_n$ preserve the bilinear form determined by the matrix with $1,-1,1,\ldots$ down the antidiagonal and zero elsewhere. So ${\mathfrak g}$ is contained in (a Lie algebra isomorphic to) $\mathfrak{so} _n$ for odd $n$, $\mathfrak{sp} _n$ for even $n$. Let's assume $n$ is odd; the argument is the same for even $n$. Clearly ${\mathfrak g}=\mathfrak{so} _3$ if $n=3$. In the general case, $[U_n, L_n]=H_n$ is a diagonal matrix with just two non-zero entries $\pm 1$ and $U_n-[H_n, U_n]$, resp. $L_n+[H_n, L_n]$ equals $U_{n-2}$, resp. $L_{n-2}$ in the "middle" matrix subalgebra. By induction, ${\mathfrak g}$ contains $\mathfrak{so} _{n-2}$. Now it's easy to see that $[H_n, U_n]$, resp. $[H_n, L_n]$ is a multiple of the "missing" positive, resp. negative root element, so in fact ${\mathfrak g}$ equals $\mathfrak{so} _n$.

$\endgroup$
1
  • $\begingroup$ This is very elegant. Thank you so much! $\endgroup$ Sep 6, 2021 at 9:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.