Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Say $A$ is a complex $n\times n$ matrix. $A$ is invertible, and moreover, supposer that the set $\{||A^n||:n=1,2,...\}$ is bounded. Prove that $A$ is diagonalizable.

My attempt was to do the following. Say $J$ is the Jordan Canonical form of $A$. Since norm is invariant under similar matrices, we have that $J^n$ is bounded as well. Let $B_1,..,B_k$ be the Jordan blocks appearing in $J$. Assume for sake of contradiction that there is a $j$ such that $B_j$ has size bigger than $1$. Then

$B_j=\lambda I +S$, where $S$ is the matrix with $1$'s in the entries above the diagonal and $0$ elsewhere.

I also know that $\lambda\neq 0$, but I dont know how to finish the problem. Any thoughts?

share|cite|improve this question
what do you mean, precisely, by $\parallel A^n \parallel ?$ Also, you should not use $n$ for both the number of rows and columns of $A$ and as the exponent. – Will Jagy Sep 21 '12 at 4:35
You should be able to work out what powers of $\lambda I+S$ look like (you could just calculate a few, and look for patterns), and see if you can deduce that the entries increase without bound. – Gerry Myerson Sep 21 '12 at 5:12
You can compute $B_j^n$ explicitly; you can take the size to be $2$ for starters. You will either find that its norm is always unbounded, or a get counterexample to what you have to prove. It seems to me the latter is actually the case, when $|\lambda|<1$. – Marc van Leeuwen Sep 21 '12 at 5:15
up vote 3 down vote accepted

This is false as stated.

Note that $$ \left( \begin{array}{rr} \frac{1}{b} & 0 \\ 0 & 1 \end{array} \right) \cdot \left( \begin{array}{rr} a & b \\ 0 & a \end{array} \right) \cdot \left( \begin{array}{rr} b & 0 \\ 0 &1 \end{array} \right) = \left( \begin{array}{rr} a & 1 \\ 0 & a \end{array} \right). $$ So that shows you how to find the Jordan form of such a 2 by 2 matrix, as soon as $b \neq 0.$

Next, take your $$ A \; = \; \left( \begin{array}{rr} \lambda & 1 \\ 0 & \lambda \end{array} \right). $$

We calculate $$ A^n \; = \; \left( \begin{array}{rr} \lambda^n & n \, \lambda^{n-1} \\ 0 & \lambda^n \end{array} \right). $$

In turn, the Jordan form of $A^n$ is $$ J_n \; = \; \left( \begin{array}{rr} \lambda^n & 1 \\ 0 & \lambda^n \end{array} \right). $$

So, if $| \lambda | \leq 1,$ the norm of $A^n$ stays bounded. Depending on the norm you mean, it may require $| \lambda | < 1$ for the norm to stay bounded.

share|cite|improve this answer
Sorry. I read the assigment again, it says $\{|A^k|: k \in \mathbb{Z}\}$, so the bound is also on the negative powers of $A$, and I know how to do it in this case. Thanks. – Daniel Montealegre Sep 21 '12 at 5:42

Take $J = \begin{bmatrix} \frac{1}{2} && 1 \\ 0 && \frac{1}{2} \end{bmatrix}$. Then a slightly tedious computation gives $J^n = \begin{bmatrix} \frac{1}{2^n} && n\frac{1}{2^{n-1}} \\ 0 && \frac{1}{2^n} \end{bmatrix}$.

Since $J^n \to 0$, it is clear that the sequence is bounded, but $J$ is not diagonalizable.

Answer to modified question (purely for my own interest):

Since $A^k v = \lambda^k v$ for an eigenvector $v$, it is clear that if $A^k$ is bounded for all integers $k$, then $|\lambda| = 1$, ie, all eigenvalues lie on the unit circle. Suppose $A$ is not diagonalizable, then $A$ has a Jordon normal form with a Jordan chain at least 2 long, ie, there exists unit vectors $v_1,v_2$ such that $Av_1 = \lambda v_1$ and $A v_2 = v_1 + \lambda v_2$. Another tedious computation shows that $A^k v_2 = k \lambda^{k-1} v_1 + \lambda^k v_2$, which contradicts $A^k$ being bounded since $\|A^k\| \geq \| A^k v_2 \| \geq k - 1$. Hence all Jordan chains are of length 1, which implies that $A$ is diagonalizable.

share|cite|improve this answer
For a less tedious computation, look at $\lambda^{-1} A$ which has the same norms and diagonalizability as $A$. Then the relation is just $Av_2 = v_1 + v_2$, which is easy to iterate. – Erick Wong Sep 21 '12 at 6:37
@ErickWong: Neat idea! – copper.hat Sep 21 '12 at 6:44

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.