I am trying to find out a way of simultaneously diagonalize multiple (larger) matrices, not necessarily by hand but with some help of Sagemath. Now I understand that for two matrices the method is as follows (correct me if I'm wrong):

For two matrices $A$ and $B$ (that commute), we can diagonalize $A$ and write it as $A=PDP^{-1}$. Then we compute $B_D=P^{-1}BP$ and compute the eigenvectors of $B_D$, say $v_1,...,v_n$. Then the vectors $w_1,...w_n$, where $w_i=Pv_i$ are eigenvectors for both $A$ and $B$.

But what if we now have a third matrix $C$, that commutes with both $A$ and $B$. Can we generalize the above procedure for multiple matrices?


That you wrote is false (take $B=I$).

Let $A_1,\cdots,A_k\in M_n$ be pairwise commuting and diagonalizable matrices with minimal polynomials $(\mu_{A_i})$. We want to simultaneously diagonalize these matrices; denote this problem by $\mathcal{P}_{n,k}$.

Step 1. For example, we assume that $A_1$ realizes the maximum of the degrees of the $\mu_{A_i}$. We diagonalize $A_1$ in the form $P^{-1}A_1P=diag(\lambda_1I_{n_1},\cdots,\lambda_pI_{n_p})$ where the $(\lambda_i)$ are distinct and $n_1+\cdots+n_p=n$. Then the $P^{-1}A_iP,i>1$ are in the form $diag(A_{i,1},\cdots,A_{i,p})$ where, for every $u\leq p$, $A_{i,u}\in M_{n_u}$ and, for every $i\not= j$, $A_{i,u}A_{j,u}=A_{j,u}A_{i,u}$.

We reduced $\mathcal{P}_{n,k}$ to the problems $\mathcal{P}_{n_1,k-1},\cdots,\mathcal{P}_{n_p,k-1}$.

Step 2. Proceed by recurrence.


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.