0

Recently, I received a "basic" linear algebra question, but it seems like I highly underestimated its difficulty. The question stated:

Consider $A,B\in M_n(\mathbb{Q})$ with $A^n=0$ but $A^{n-1}\neq 0$. Show \begin{align*} AB=BA \implies B= F(A)\text{ for some }F \in \mathbb{Q}[x]\text{ such that }\operatorname{deg}(F)<n \end{align*}

This question is basically asking us to prove the centralizer of $A$ must be $\mathbb{Q}[A]$, which I guess now is highly nontrivial. One of my attempts is to argue that the dimension of the centralizer of $A$ is $n$, since the Jordan form of $A$ is pretty nice, but I fail to prove how commutativity is preserved under conjugation, and I also fail to compute the centralizer of Jordan block. Anyway, that was just my attempt. It may now really work, so if you have another idea, it is very likely yours is better.

Thanks for any help in any form in advance.

  • $(PAP^{-1})(PBP^{-1})=PABP^{-1}$, but perhaps you're more interested in $F(PAP^{-1})=PF(A)P^{-1}$. More generally, conjugation by an invertible matrix is an isomorphism of between subalgebras of $M_n(\Bbb Q)$. – Sassatelli Giulio Jun 03 '25 at 12:45
  • 4
    It's just a matter of writing down what matrices commute with a Jordan nilpotent block. You will find that they coincide with the upper triangular Toeplitz matrices, that are exactly the linear combinations of the powers of the Jordan block – Exodd Jun 03 '25 at 12:50
  • You may be interested in this (vast) generalization of your statement, characterizing exactly when $C_{M_n(K)}(A) = K[A]$ https://math.stackexchange.com/a/92832/1213739 – Leobeth Jun 03 '25 at 13:10
  • This exact question regarding order n nilpotent matrices has been asked many times on this site, e.g. in the last few months: here https://math.stackexchange.com/questions/5052327/prove-that-g-commutes-with-f-if-and-only-if-g-in-textspan-big-tex and https://math.stackexchange.com/questions/5012778/proving-the-existence-of-a-polynomial-pf-for-g-when-f-is-nilpotent-with – user8675309 Jun 03 '25 at 14:16

1 Answers1

1

Since $A^n=0$ but $A^{n-1}\neq0$, the minimal polynomial of $A$ must be $x^n$. In particular, no smaller power of $A$ vanishes. Because the characteristic polynomial also has degree $n$, it too must be $x^n$. In other words, $A$ is nilpotent of index exactly $n$, and its minimal and characteristic polynomials coincide—so $A$ is “non-derogatory.”

Equivalently, over $\mathbb{Q}$ one can put $A$ into a single $n\times n$ Jordan block for the eigenvalue $0$. Concretely, there is an invertible $P\in GL_n(\mathbb{Q})$ so that

$$ A \;=\; P\,J\,P^{-1}, \quad J \;=\; \begin{pmatrix} 0 & 1 & 0 & \cdots & 0\\ 0 & 0 & 1 & \cdots & 0\\ \vdots & & \ddots & \ddots & \vdots\\ 0 & 0 & \cdots & 0 & 1\\ 0 & 0 & \cdots & 0 & 0 \end{pmatrix}. $$

Because $AB=BA$, if we set $B' = P^{-1}BP$, then $J\,B' = B'\,J$. Writing $B'=(b_{ij})$, one checks that this forces

$$ b_{i,j-1} \;=\; b_{\,i+1,j} \quad \text{and}\quad b_{k1}=0 \text{ for }k>1,\quad b_{n,j-1}=0\text{ for }j>1. $$

In plain language, each super-diagonal of $B'$ has a constant entry, and every entry strictly below the main diagonal must be zero. Thus

$$ B' \;=\; \begin{pmatrix} c_0 & c_1 & c_2 & \cdots & c_{n-1}\\ 0 & c_0 & c_1 & \cdots & c_{n-2}\\ 0 & 0 & c_0 & \cdots & c_{n-3}\\ \vdots & \vdots & & \ddots & \vdots\\ 0 & 0 & 0 & \cdots & c_0 \end{pmatrix}, $$

with each $c_k\in\mathbb{Q}$. But that is exactly

$$ c_0 I + c_1 J + c_2 J^2 + \cdots + c_{n-1} J^{\,n-1}. $$

Since $J^n=0$, the polynomial

$$ F(x) \;=\; c_0 + c_1 x + \cdots + c_{n-1}x^{\,n-1} $$

satisfies $B' = F(J)$. Going back to the original basis,

$$ B \;=\; P\,B'\,P^{-1} \;=\; P\,F(J)\,P^{-1}. $$

But $J = P^{-1}AP$, so $F(J) = P^{-1}F(A)\,P$. Hence

$$ B \;=\; F(A), $$

and $F$ has rational coefficients of degree at most $n-1$.

Once you realize $A$ is a single $n$-block Jordan nilpotent, any matrix commuting with it must be a polynomial in $J$; conjugating back shows $B$ is that same polynomial in $A$. Therefore

$$ B \;=\; F(A), \quad F(x)\in \mathbb{Q}[x],\;\deg F < n, $$

as claimed.

John
  • 491
  • 1
    Decomposition into Jordan form requires the ground field to be closed. How did you deduce that all entries of $P$ are rational? – John Frank Jun 03 '25 at 13:09
  • @JohnFrank Just take $X_1\in\mathrm{Im}(A^{n-1})\setminus{0}$ and define $X_n$ such that $A^{n-1}X_n=X_1$. Then $X_k=A^{n-k}X_n$ and the base $(X_1,\dots,X_n)$ provides the Jordan form. – Christophe Boilley Jun 03 '25 at 13:24
  • @JohnFrank No. A matrix over a field $F$ is similar to its Jordan form over $F$ if and only if it has a full spectrum in $F$. In particular, every nilpotent matrix is similar to a nilpotent Jordan block over the ground field, because the field always contains the zero element. – user1551 Jun 03 '25 at 16:35