# How to compute a simultaneous block-diagonalization?

Let $$n$$ be a positive integer and consider of finite set $$S \subset M_n(\mathbb{C})$$ such that $$S^* = S$$ (i.e. if $$a \in S$$ then $$a^* \in S$$). The algebra generated by $$S$$ is a finite dimensional $$*$$-algebra over $$\mathbb{C}$$, so is isomorphic to a direct sum of matrix algebras, i.e. there are $$n_1 \le n_2 \le \dots \le n_r$$ such that:
$$\langle S \rangle \simeq \bigoplus_{i=1}^r M_{n_i}(\mathbb{C})$$

Question: How to compute the change of basis $$p$$ such that for all $$a \in S$$, we have $$p^{-1}ap$$ block-diagonal as for the above decomposition?

We can see each $$M_{n_i}(\mathbb{C})$$ above as $$End(V_i)$$ with $$V_i$$ an irreducible representation of $$\langle S \rangle$$.

If the matrices commute over each other then (using $$S^*=S$$) we get that the matrices are normal, so diagonalizable, and then what we ask in this case is just a simultaneous diagonalization (ok). So in general, what I am looking for is how to compute a (thinnest) simultaneous block-diagonalization.

• Vrej Zarikian wrote a preprint that solves this problem (and a couple others) called "Algorithms for Operator Algebra Calculations" in 2003. However, I am not able to find a public copy of this paper anywhere online, so I am hesitant to share it without permission. Maybe you could e-mail him and ask? usna.edu/Users/math/zarikian/index.php Sep 17, 2019 at 11:42