Let $G$ be a finite group and $\rho : G\to GL(V)$ a representation of $G$ on the vector space $V$. If $g\in G$ is one element of order $n$ we have $g^n=1$, so that


where $I$ is the identity linear transformation. Now, if we pick a basis we can convert $\rho(g)$ to a matrix. In particular, we can pick the Jordan basis of $\rho(g)$ so that the matrix $A$ of $\rho(g)$ takes the block-diagonal form

$$\begin{pmatrix}J_1 & 0 & \cdots & 0 \\0 & J_2 & \cdots & 0 \\ \vdots & \vdots & \ddots & 0 \\ 0 & \cdots & \cdots & J_k \end{pmatrix}$$

where $J_i$ is a Jordan matrix, that is, for each $i\in \{1,\dots,k\}$ there is $\lambda_i$ such that $J_i$ is diagonal with $\lambda_i$ on the diagonal and with $1$'s right above.

I've heard that this implies that the action of $\rho(g)$ is diagonal with the $n$-th roots of unity on the diagonal.

I must confess that I'm not being able to understand this, though.

  1. How do we derive this result? I believe it has to do with the fact that the minimal polynomial of $\rho(g)$ divides $x^n-1$, but I don't know how to use this.

  2. For a given $n$ the number of $n$-th roots of unity is $n$. If I understood $A$ would have $n$ elements on the diagonal, implying that $A$ represents a linear map on one $n$-dimensional space. But $n$ is just the order of $g$, the representation $V$ is arbitrary! It could have dimension greater than $n$ or less than $n$. How can this be?

So how to understand those two issues? How to show that $A$ is diagonal with the $n$-th roots of unity on the diagonal and how to reconcile this with the fact that $V$ can have any dimension?


Let $ K $ be a field containing the $ k $th roots of unity such that its characteristic does not divide $ k $, and let $ A $ be a nilpotent matrix in $ GL_n(K) $ with $ A^k = I $. $ A $ is a root of $ x^k - 1 $, which splits into distinct linear factors in $ K[x] $. The minimal polynomial of $ A $ then also splits into distinct linear factors, and therefore we have that $ A $ is diagonalizable. Its eigenvalues are roots of its minimal polynomial which divides $ x^k - 1 $, therefore they are all $ k $th roots of unity.


Assume that $V$ is a vector space over a field $K$.

Supplementing Starfall's answer to your first question with the following facts about diagonalizability of $A$.

If $\operatorname{char} K=0$, then you can argue as follows. Any Jordan block $J$ necessarily also satisfies $J^n=I$. But if $J$ has more than one row (and hence at least a single $1$ adjacent to the diagonal), then an easy induction shows that the diagonal entries of $J^m$ are all equal to $\lambda^m$, and the adjacent entries are all equal to $m\lambda^{m-1}$. So when $\lambda$ is a root of unity, and $m\cdot 1_K\neq0_K$ those non-diagonal entries are non-zero. Therefore all the Jordan blocks have size one and we are done.

On the other hand, if $\operatorname{char} K=p>0$, then the above argument fails. Actually the claim is also false. An element $g$ of order $p$ may be represented by the matrix $$ \rho(g)=\pmatrix{1&1\cr0&1\cr} $$ when we actually have $$ \rho(g)^p=\pmatrix{1&p\cr0&1\cr}=I_2. $$

Re: Your second question. Different Jordan blocks can share the same $\lambda$, and it is also possible that some roots of unity do not show up at all in the Jordan decomposition of $A$. So the number of distinct $n$th roots of unity cannot be compared with the size of the matrix $A$ at all.

  • 1
    $\begingroup$ Of course, in your counterexample the exponent is divisible by the characteristic of your field; which makes it so that the minimal polynomial actually becomes inseparable. Still, it is interesting to note that such a counterexample is possible precisely when the characteristic of the field divides the order of $ G $. $\endgroup$ – Starfall Jun 7 '16 at 7:55
  • $\begingroup$ @Starfall: That's right! A related fact is that in characteristic $p$ there are no roots of unity of order $p$. So by requiring that the field $K$ contains the necessary roots of unity we can actually exclude this possibility :-) $\endgroup$ – Jyrki Lahtonen Jun 7 '16 at 8:08

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.