Suppose that $A,B\in M_n(\mathbb{C})$ such that $AB-BA=A $. Prove that $A$ is not invertible.

My work:

Suppose $A$ is invertible. Then $ABA^{-1}=I+B$ . So $B$ is similar to $I+B$ .Let $B$ have eigenvalues $c_1,c_2,\ldots,c_n \in \mathbb{C}$. So $B$ has basis such that $B$ is upper triangular with respect to it and has $c_1,\ldots,c_n$ as diagonal entries . It is easy to see that $I+B$ is upper-triangular with respect to this basis and has entries $1+c_1,\ldots, 1+c_n$ .

Hence $$c_1+c_2+\ldots +c_n=\operatorname{trace}(B)$$ $$=\operatorname{trace}(I+B)=(1+c_1)+\ldots+(1+c_n)=n+c_1+\ldots +c_n .$$

So $n=0$, contradiction. I'm not sure if my solution is correct one. It seems alright. I will be very thankful if you can confirm that the proof indeed is a correct one.

Any other possible solutions are welcomed.

  • 16
    $\begingroup$ It is correct, but too complicated. After proving that $B$ and $I+B$ are similar, you can simply say that this is no possible, since the trace of $I+B$ is $n$ plus the trace of $B$. No need to mention upper triangular matrices. $\endgroup$ May 28, 2017 at 16:38

3 Answers 3


Let $A$ be invertible.

We have $(AB-BA)A^{-1}=I$, or $ABA^{-1}-B=I.$

But $\operatorname{tr} (ABA^{-1})=\operatorname{tr}{B},$ which is a contradiction.


It is probably worth showing that $A$ is in fact nilpotent. Indeed, by induction we get $$A^m B - B A^m = m A^m$$ for all $m\ge 0$. Taking the traces on both sides we get $$0=m \operatorname{Trace}A^m$$ so (assuming char $0$) $$\operatorname{Trace}A^m = 0$$ for all $m\ge 1$. This implies $A$ nilponent.

$\bf{Added:}$ Based on an idea of @Hans: , we can generalize this . For $X$, $Y$ matrices, let $[X,Y]=X Y - Y X$.

Assume that $[C,B]=A$, and $[C,A]=0 $ ( $C$, $A$ commute). Then $A$ nilpotent.

Indeed, for all $m\ge 1$ we have $[C,A^{m-1} B] = A^{m-1} [C,B]= A^m$, and so $\operatorname{Trace} A^m=0$ . Conclude $A$ nilpotent (assume char $0$).

$\bf{Added:}$. What we are showing is that if $\operatorname{ad}(C)^2 A=0$, then $\operatorname{ad}(C) A$ is nilpotent. We can also reason as follows: we may assume that the basic field is algebraically closed. Consider the Jordan structure of $C$, with blocks of size $n_1$, $\ldots$, $n_k$ corresponding to $\lambda_1$, $\ldots$, $\lambda_k$. Then we see that all operators $B$ in the kernel of $\operatorname{ad}C^2$ have the property that $\operatorname{ad}(C)B=0$.

(here we use an explicit Jordan structure for $\operatorname{ad}(C)$. We again use that the characteristic of the field is $0$ or $>n$ ).

  • 1
    $\begingroup$ Nice conclusion. +1. You only need to show $A^mB-A^{m-1}BA=A^m\implies \text{tr}A^m=0, \forall m\in\mathbf N$. $\endgroup$
    – Hans
    Oct 8, 2017 at 7:34
  • $\begingroup$ @Hans; yes, your approach is simpler and better! I should add this idea. $\endgroup$
    – orangeskid
    Oct 8, 2017 at 17:46

$\def\m{\mathfrak }$This result (and, in fact, orangeskid's observation) has the following far reaching generalization:

if $\m g$ is a finite dimensional (complex) Lie algebra and $\mathfrak r$ is its radical, then $[\m g,\m r]$ acts on any finite dimensional representation of $\m g$ nilpotently.

Indeed, in the question the matrices $A$ and $B$ span a Lie algebra $\m g$ of dimension (at most, really, but let us suppose) equal to $2$ which is solvable, so that the radical of $\m g$ is simply $\m g$ itself, and the theorem above tells us that $[\m g,\m g]$ acts nilpotently on finite dimensinal modules: since $[\m g,\m g]$ is spanned by $A$, the desired result follows.

Of course, this is immensely more general.

As an example:

if $A$, $B$ and $C$ are square matrices such that $CA-AC = B$, $CB-BC = aA + B$, $AB-BA = 0$, then $B$ is nilpotent and, more generally, al linear combinations of $A$ and $B$.

This comes from looking at a random 3-dimensional solvable Lie algebra.

  • $\begingroup$ One can find the proof of this, which depends on Levi's theorem, in the book on Lie algebra by Hossein Abbaspour and Martin A. Moskowitz; it is Theorem 3.5.6 there. $\endgroup$ Oct 12, 2017 at 17:41
  • $\begingroup$ Excellent answer! I will have to review my rusty rep theory... $\endgroup$
    – orangeskid
    Oct 12, 2017 at 19:16

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.