I have to make an argument that the characteristic polynomial of an idempotent matrix A is precisely (maybe a constant multiple of) $\lambda^{n-r}(1-\lambda)^{r}$, where r is the rank of A. I know that for an idempotent matrix A, the eigenvalue is either 0 or 1 and that the 1-eigenspace of A is the image of A. Furthermore, I was given the hint to use the geometric multiplicity of the eigenvalue 1 and to entertain the questions: What is the geometric multiplicity of 0 as an eigenvalue? What is the sum of these two multiplicities? Is it possible that either eigenspace is deficient?

Update: Following Steven's suggestions, I tackled the problem as follows:

  1. I found that geometric multiplicities as follows: $$ \begin{align*} gmu(0)&=dim(ker(A-(0)I)) \\ &=dim(ker(A)) \\ &=n-rank(A-I) \\ &=n-r \end{align*} $$ and $$ \begin{align*} gmu(1)&=dim(ker(A-(1)I)) \\ &=dim(E_{1}) \\ &=dim(im(A)) \\ &=n \end{align*} $$

  2. I then use these multiplicities to argue that the minimal polynomial must be of the form: $$ f_{A}(\lambda)=(\lambda)^{n-r}(1-\lambda)^{r} $$ where the roots are either 0 or 1, as suggested by the answer below. Then, matrix A is diagonalizable and similar to a diagonal matrix D with r $1$'s and (n-r) $0$'s as its diagonal entries.

  • 2
    $\begingroup$ The rank-nullity theorem should be of use here $\endgroup$ May 12, 2021 at 9:36

1 Answer 1


Question: "How can one approach this argument? Any help would be appreciated."

Answer: Let $A$ be an $n \times n$ matrix with $A^2=A$. Let $v \neq 0$ be an eigenvector with eigenvalue $\lambda$. It follows $Av=\lambda v$ and

$$\lambda v=Av=A^2v=A(Av)=A(\lambda v)=\lambda^2v$$

hence $\lambda^2=\lambda$ and hence $\lambda =0,1$. Hence the eigenvalues of $A$ are $0,1$. It follows the characteristic polynomial (which has degree $n$) must look as follows:


for some integer $r$, since the eigenvalues of $A$ equal the roots ot $\chi_A(t)$.

  • $\begingroup$ Hi, @hm2020. I follow you all the way through the following statement: "It follows the characteristic polynomial (which has degree 𝑛) must look as follows." I am failing to see the connection of that with the previous statement that lambda = 0 or 1. Could you provide a further explanation as to why it must be of that form (which is not the form that I'm used to seeing: e.g., $t^{2}-tr(A)t+det(A)$ for a 2 by 2 )? Am I missing something important conceptually or algebriaclly here? Thanks. $\endgroup$
    – Yang Wu
    May 12, 2021 at 18:02

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.