Every $n\times n$ matrix will produce an $n$-th order characteristic equation. This polynomial will have $n$ distinct roots or less than $n$ distinct roots (with some repeated).

In the former case, each eigenvalue corresponds to a distinct eigenvector, so you can form an eigenbasis of $n$ linearly independent eigenvectors.

In the latter case, I can see two possibilities:

  1. There are still $n$ linearly independent eigenvectors, but some share the same eigenvalue, so a $k$-fold root of the characteristic equation (an eigenvalue repeated $k$ times) will correspond to $k$ linearly independent eigenvectors.

  2. There are less than $n$ linearly independent eigenvectors, so a $k$-fold root may correspond to less than $k$ linearly independent eigenvectors.

Is one of these two possibilities always true, for all complex $n\times n$ matrices, or could either possibility be true depending on the matrix in question?

Is there a proof of this, or perhaps a counter example?

  • $\begingroup$ $\begin{bmatrix}1&1\\0&1\end{bmatrix}$ $\endgroup$ Dec 24, 2018 at 17:23

2 Answers 2


Both possibilities apply. Case (1) means the matrix is diagonalisable over $\mathbb{C}$, of which there clearly are examples. However matrices like $$\begin{bmatrix}1 & 1 \\ 0 & 1\end{bmatrix}$$ are ones that are not diagonalisable over $\mathbb{C}$; this one would fit your case (2).

The reason why we would consider eigenvalues over $\mathbb{C}$ rather than over $\mathbb{R}$ is precisely because the characteristic polynomial will split as a products of its roots, which does not always happen over $\mathbb{R}$. Hence we always find (counting multiplicities) $n$ eigenvalues, but it turns out this is not sufficient for diagonalisability.

  • $\begingroup$ Thanks. I've also actually just managed to find $\begin{bmatrix} 0 & 1 & 1 \\ 1 & 0 & 1 \\ 1 & 1 & 0 \\ \end{bmatrix}$ as an example of a matrix with one distinct eigenvalue and one degenerate eigenvalue, but three linearly independent eigenvectors. $\endgroup$ Dec 24, 2018 at 17:38
  • 1
    $\begingroup$ @Pancake_Senpai If you want to look into this stuff in more detail, I suggest learning something about the Jordan canonical form of a matrix. The Jordan canonical form looks to the eye to almost be diagonalisation of it. Every matrix can be brought into a Jordan canonical form, even though not all matrices are diagonalisable. Moreover, you can read off from the Jordan canonical form whether a matrix is diagonalisable or not. $\endgroup$
    – SvanN
    Dec 24, 2018 at 18:49
  • $\begingroup$ Thanks for the info! $\endgroup$ Dec 25, 2018 at 0:30

Standard example: $$ \pmatrix{0 & 1\cr 0 & 0\cr}$$ with only one linearly independent eigenvector.


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.