Let $A \in GL_n(\mathbb R)$ be a diagonalizable matrix. Let $C_A = \{B \in M_n(\mathbb R): AB=BA\}$ denote the set of centralizers of $A$. Let $\phi: M_n(\mathbb R) \to \mathbb R^n$ be a linear map defined by $C \mapsto (CA-AC)e_1$ where $e_1 = (1, 0, \dots, 0)^T$. Let $M_A = \text{ker}(\phi) = \{B \in M_n(\mathbb R): (BA-AB)e_1 = 0\}$. Then $C_A \subseteq M_A$. Is it possible that $C_A$ is dense in $M_A$?


The dimension of $M_A := \ker(\phi)$ is at least $n^2-n$ by rank-nullity theorem applied to $\varphi: M_{n}(\mathbb{R})\to \mathbb{R}^n$. On the other hand, if $A$ is diagonalizable with eigenvalues $\lambda_1$, ..., $\lambda_r$ with respective multiplicities $n_1, ..., n_r$ (so $n_1 + n_2 + ... + n_r = n$), then the dimension of the centralizer: $$ \dim C_A = n_1^2 + n_2^2 + ... + n_{r}^2 $$ This can be proved from the observation that any matrix $B$ which commutes with $A$ must leave each eigenspace of $A$ invariant (so $B$ can be represented as a block-diagonal matrix with $r$ blocks of size $n_1, n_2, ..., n_r$, and it is clear that the number of such matrices is $n_1^2+n_2^2+...+n_r^2$).

If $r=1$, then $A$ is a scalar multiple of identity and the problem is easy to solve -- in this case $C_A=M_n(\mathbb{R})$. Suppose $r>1$. Then there at least 2 distinct eigenvalues, and $$ n_1^2+n_2^2+...+n_r^2 \leq (n-1)^2+1^2 = n^2 - 2n + 2 < n^2-n $$ for $n>2$. So $\dim(C_A) < \dim(M_A)$, and hence $C_A$ cannot be dense in $M_A$ (because it has a lower dimension).

When $n=2$ and $r=2$, then $\dim(C_A)=1^2+1^2 = 2 = n^2 - n\leq \dim(M_A)$, so there is a chance that the dimensions could coincide. I bet if you take a generic enough $2\times 2$ diagonalizable matrix (which will have distinct eigenvalues), then $\dim(M_A)$ will equal to $n^2-n=2$ which coincides with dimension of $C_A$, so in that case $C_A=M_A$.

| cite | improve this answer | |

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.