The Unapologetic Mathematician

Mathematics for the interested outsider

Cartan Matrices

As we move towards our goal of classifying root systems, we find new ways of encoding the information contained in a root system \Phi. First comes the Cartan matrix.

Pick a base \Delta\subseteq\Phi of the root system. Since V is finite-dimensional and \Delta is a basis of V, \Delta must be finite, and so there’s no difficulty in picking some fixed order on the simple roots. That is, we write \Delta=\{\alpha_1,\alpha_2,\dots,\alpha_n\} where n=\dim(V). Now we can define the “Cartan matrix” as the n\times n matrix whose entry in the ith row and jth column is \alpha_i\rtimes\alpha_j. These entries are called “Cartan integers”.

The matrix we get, depends on the particular ordering of the base we chose, of course, so the Cartan matrix isn’t quite uniquely determined by the root system. This is relatively unimportant, actually. More to the point is the other direction: the Cartan matrix determines the root system up to isomorphism!

That is, let’s say \Delta'\subseteq\Phi'\subseteq V' is another root system \Phi' in another vector space V with another identified base \Delta'=\{\alpha'_1,\dots,\alpha'_n\}. Further, assume that for all 1\leq i,j\leq n we have \alpha'_i\rtimes\alpha'_j=\alpha_i\rtimes\alpha_j, so the Cartan matrix determined by \Delta' is equal to the Cartan matrix determined by \Delta. I say that the bijection \alpha_i\mapsto\alpha'_i extends to an isomorphism \phi:V\rightarrow V' that sends \Phi onto \Phi' and satisfies \phi(\alpha)\rtimes\phi(\beta)=\alpha\rtimes\beta for all roots \alpha,\beta\in\Phi.

The unique extension to \phi is trivial. Indeed, since \Delta is a basis for V all we have to do is specify all the images \phi(\alpha_i) and there is a unique linear transformation \phi:V\rightarrow V' extending the mapping on basis vectors. And it’s an isomorphism, since the image of our basis of V is itself a basis of V', so we can turn around and reverse everything.

Now our hypothesis that the bases give rise to the same Cartan matrix allows us to calculate for simple roots \alpha,\beta\in\Delta:

\displaystyle\begin{aligned}\sigma_{\phi(\alpha)}(\phi(\beta))&=\phi(\beta)-(\phi(\beta)-\phi(\alpha)\phi(\alpha)\\&=\phi(\beta)-(\beta\rtimes\alpha)\phi(\alpha)\\&=\phi(\beta-(\beta\rtimes\alpha)\alpha)\\&=\phi(\sigma_\alpha(\beta))\end{aligned}

That is, \phi intertwines the actions of each of the simple reflections \sigma_\alpha. But we know that the simple reflections with respect to any given base generate the Weyl group!

And so \phi must intertwine the actions of the Weyl groups \mathcal{W} and \mathcal{W}'. That is, the mapping \sigma\mapsto\phi\circ\sigma\circ\phi^{-1} is an isomorphism \mathcal{W}\rightarrow\mathcal{W}' which sends \sigma_\alpha to \sigma_{\phi(\alpha)} for all \alpha\in\Delta.

We can go further. Each root \beta is in the \mathcal{W}-orbit of some simple root \alpha. Say \beta=\sigma(\alpha) for \alpha\in\Delta. Then we find

\displaystyle\phi(\beta)=\phi(\sigma(\alpha))=(\phi\circ\sigma\circ\phi^{-1})(\phi(\alpha))\in\Phi'

And so \phi must send \Phi to \Phi'. A straightforward calculation (unwinding the one before) shows that \phi must then preserve the Cartan integers for any roots \alpha and \beta.

February 16, 2010 Posted by | Geometry, Root Systems | 8 Comments

   

Follow

Get every new post delivered to your Inbox.

Join 392 other followers