Let $\mathfrak g$ be a semisimple Lie algebra over $\mathbb{C}$ with Cartan subalgebra $\mathfrak h$. Then the Killing form on $\mathfrak g$ defined by $(X,Y) = \operatorname{Tr}(\operatorname{ad}(X) \circ \operatorname{ad}(Y))$ is nondegenerate (this can also be taken as the definition of a semisimple Lie algebra).

I'm trying to understand the following:

1 . Why is the restriction of the Killing form to $\mathfrak h$ still nondegenerate?

2 . Why do we have $(h,h) \neq 0$ for all $h \in \mathfrak h$?

3 . Let $\Phi \subseteq \mathfrak h^{\ast}$ be the roots of $\mathfrak h$ in $\mathfrak g$. Let $V$ be the $\mathbb{Q}$-span of $\Phi$, and define a bilinear form on $V$ by $(v,w) = (h_v,h_w)$, where $h_v \in \mathfrak h$ is the inverse of $v$ under the $\mathbb{C}$-vector space isomorphism $ \mathfrak h \rightarrow \mathfrak h^{\ast}, h \mapsto (-,h)$. Why does this define a symmetric, positive definite (hence nondegenerate) bilinear form on $V$?

The third statement is a claim made on page 1 of Steinberg, Lectures on Chevalley Groups.

Proof of 1: the linear transformations $\operatorname{ad}(h) : h \in \mathfrak h$ are simultaneously diagonalizable, with diagonal matrix $\operatorname{diag}(0, \ldots , 0, \alpha(h) : \alpha \in \Phi)$, where the number of zeros is the dimension of $\mathfrak h$. So the Killing form on $\mathfrak h$ is

$$(h,h') = \sum\limits_{\alpha \in \Phi} \alpha(h)\alpha(h') = 2\sum\limits_{\alpha \in \Phi^+} \alpha(h)\alpha(h')$$

where $\Phi^+$ is a given set of positive roots with base $\Delta$. Let $h_1, \ldots , h_t$ be a dual basis to simple roots $\alpha_1, \ldots , \alpha_t$. Fix $1 \leq i \leq t$, and each $n \geq 1$, let $m_n$ be the number of positive roots with the coefficient $n\alpha_i$. Then

$$(h,h_i) = 2(m_1 + 2m_2 + \cdots)\alpha_i(h)$$

So if $(h,h') = 0$ for all $h' \in \mathfrak h$, then $\alpha_i(h) = 0$ for all $i$ $\Rightarrow h = 0$.

  • 1
    $\begingroup$ When you write \textrm{Tr} instead of \operatorname{Tr} then you don't automatically get proper spacing in things like $A\operatorname{Tr} B$ and $A\operatorname{Tr}(B),$ instead seeing $A\textrm{Tr} B$ and $A\textrm{Tr}(B).$ I include both examples to show the context-dependent nature of the spacing: there's less space when $\Big($parentheses$\Big)$ are there. In LaTeX, before \begin{document} you can write \newcommand{\d}{\operatorname{diag}} and then just use \d subsequently in the document and see $\operatorname{diag} in the finished result. $\endgroup$ – Michael Hardy Dec 24 '17 at 1:03
  • $\begingroup$ This is proved, among many other places, in Carter's book on Lie algebras. Have you looked in any of the standard textbooks on the subject? $\endgroup$ – Mariano Suárez-Álvarez Dec 24 '17 at 2:28
  • 1
    $\begingroup$ Maybe I'm confused but did you just give a proof of the statement you asked about? $\endgroup$ – Tim kinsella Dec 24 '17 at 6:46
  • $\begingroup$ Only the first one $\endgroup$ – D_S Dec 24 '17 at 6:59
  • $\begingroup$ The other two statements follow from the fact the formula you have for $(h, h')$ and the fact that the roots are real valued on $V^\ast$ $\endgroup$ – Tim kinsella Dec 25 '17 at 23:53

You have proved the first statement as part of the question. The second statement is not true and cannot be true: The Killing form is a complex bilinear form, and such a form always has isotropic vectors. you want to state this only for the subspace $\mathfrak h_0\subset \mathfrak h$ on which all roots have real values. It is part of the analysis of the root decompostion to prove that this is positive definite and this is a slightly involved sequence of arguments that you can find in standard text. The third part is a slight variation of this.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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