Let a set of elements, $T^i_j$, with $i,j=1,\cdots,n$ satisfying the $\mathcal{su}(n)$ algebra $$ [T^i_j, T^k_l] = \delta^k_j T^i_l - \delta^i_l T^k_j\,,\qquad (T^i_j)^\dagger = T^j_i. $$ There are $n^2$ elements, but It's easy to see that the "trace" commutes with everything, so we can remove an abelian factor by shifting $T^i_j\to T^i_j-\frac{1}{n}\delta^i_j \sum_k T^k_k$, without changing the commutation relations.

I'm asked to find the Cartan matrix for $n=4$ starting from the commutation relation above. Adapting this question I found that the Cartan subalgebra is given by $\mathcal{H}=\{T^1_1- T^2_2, T^2_2-T^3_3, T^3_3-T^4_4\}$. I can then find the weights from

$$ [H_a, E_\alpha] = \alpha^a E_\alpha\,,\qquad H_a\in\mathcal{H} $$ Going through all the non-Cartan generators to find their roots $\alpha=(\alpha^1,\alpha^2,\alpha^3)$, I managed to find those which have the same values as the Cartan matrix of $\mathfrak{su}(4)$, and are also simple roots (all the other roots can be obtain from these): $$ E_{\alpha_1}=T^1_2:\qquad~ \alpha_1 =(2,-1,0)\\ E_{\alpha_2}=T^2_3:\qquad~~~ \alpha_2 =(-1,2,-1)\\ E_{\alpha_3}=T^3_4:\qquad \alpha_3 =(0,-1,2) $$ How can I actually compute the Cartan Matrix from there? I would like to use the usual formula $$ A_{ij} = 2 \frac{(\alpha_i,\alpha_j)}{(\alpha_i,\alpha_i)}\stackrel{\mathfrak{su}(4)}{=} \begin{pmatrix} 2&-1&0\\ -1&2&-1\\ 0&-1&2 \end{pmatrix} $$ But I don't know how to define the pairing $(\cdot,\cdot)$ on the root lattice in this case. I cannot use the Cartesian product, because that would give the wrong result.

Usually positive roots are defined as having the first non-vanishing entry positive, which is not the case for the third one. Contrary to the question mentioned above, I am given only the commutation relation and not the form of the generators, i.e. I don't have $(T^i_j)_{ab}= ...$ so I cannot define the usually Killing form $\left<T_a, T_b\right>\propto\delta_{ab}$ to map the algebra to the root lattice. Moreover in the usual construction we have $$ [E_\alpha, E^\dagger_\alpha] = \sum_i \alpha_i H_i $$ which is not the case here. Is there a canonical way to find the Cartan matrix in this case?

  • $\begingroup$ There are many different notations for these things, but I do not know what $\alpha = (2,-1,0)$ etc. is supposed to mean. Please explain, how is that a root of $su(3)$? $\endgroup$ Oct 16, 2020 at 16:26
  • $\begingroup$ I meant $\mathfrak{su}(4)$, and instead wrote its rank. I edited the question and added some additional definitions for clarity. $\endgroup$
    – Bulkilol
    Oct 16, 2020 at 21:29

1 Answer 1


The relation (I switch $a$ to $i$ to make it look more distinguishable from $\alpha$) $$ [H_i, E_\alpha] = \alpha^i E_\alpha\,,\qquad H_a\in\mathcal{H} $$

would more commonly be written


But now if $H_i$ is the coroot to the root $\beta_i$ (i.e. $H_i$ is the unique element of $[E_{\beta_i}, E_{-\beta_i}]$ for which $\beta_i(H_i)=2$) then

$$\alpha(H_i)= \check{\beta_i}(\alpha)$$

and it's one of the first things shown in any serious introduction to root systems that, if $( \cdot, \cdot)$ is a bilinear form on the (vector space ambient to the) root system which is invariant under root system automorphisms, then

$$\check{\beta}(x) = \dfrac{2 (\beta, x)}{(\beta, \beta)}.$$

Putting everything together, you have

$$\alpha^i = \dfrac{2 (\beta_i, \alpha)}{(\beta_i, \beta_i)}$$

or, if I understand your notation correctly,

$$\alpha_j^i = \dfrac{2 (\alpha_i, \alpha_j)}{(\alpha_i, \alpha_i)}.$$

So there you have the Cartan matrix as the transpose of what you get when you write your $\alpha_i$ underneath each other (in this case, the transpose doesn't do anything anyway).

The upshot being that if you already know the numbers you call $\alpha^a$, you do not need to define the form $(\cdot, \cdot)$ -- everything you need to know about the Cartan matrix is in those numbers. (And they actually, "the other way around", define such a form $(\cdot, \cdot)$ uniquely up to scaling.)

If you insist on having a form $(\cdot, \cdot)$ which comes from the Lie algebra you have and not through the technicalities of the root system: Try the Killing form, but be careful, because a priori that one is defined on (e.g.) elements of the Cartan subalgebra, which are coroots, so some dualising might be necessary which in a given example might or might not change some numbers.

Finally, I'd like to point out that it seems everything we're doing here is not happening inside $\mathfrak{su}(n)$ literally, but rather its complexification which is $\simeq \mathfrak{sl}_n(\mathbb C)$ (otherwise, there are no roots and root spaces $E_\alpha$). Also, there is not "the" Cartan subalgebra: Every non-zero semisimple Lie algebra has infinitely many Cartan subalgebras, it's just that usually the diagonal matrices or some variant thereof are the most convenient one.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .