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

This is from Berkeley Problems in Mathematics, Spring 1979:

Let $M$ be a $3\times 3$ non-singular real matrix. Prove that there are real matrices $S$ and $U$ such that $M=SU=US$, all the eigenvalues of $U$ equal 1, and $S$ is diagonalizable over $\mathbb{C}$.

I am thinking that since $U$'s eigenvalues are all 1, we have $(U-I)^{3}=0$. Thus up to Jordon canonical form $U$ should be either a whole Jordan block or has a rank 2 Jordan block. But I do not know how to use this to approach this problem. On the other hand $S$ can be expressed in $PDP^{-1}$, and we are supposed to solve $A=PDP^{-1}U=UPDP^{-1}$. Again no familiar formula can be obtained this way. So I venture to ask for a hint. This may be related to Schur decomposition but I do not know it well enough to solve it.

share|cite|improve this question
The original Problem 7.7.13 (Sp79) requires $M$, $S$, $U$ real and $M$ non singular: – enzotib Aug 1 '12 at 17:38
sorry for the typo. shall fix it. – Bombyx mori Aug 1 '12 at 17:48
I feel you can use the polar decomposition of a matrix, but could not say exactly how. – enzotib Aug 1 '12 at 18:26
The main catch (for me) is the real requirement. – copper.hat Aug 1 '12 at 18:48
up vote 2 down vote accepted

Here is a rather tedious proof:

Since $M$ is real, either the eigenvalues are all real, or there is one real and a pair of conjugate eigenvalues.

The key result here is that a real matrix has a real Jordan form, not quite upper triangular, but 'almost'. In the invertible $3 \times 3$ case, this means that there exists a real $V$ such that $J = V^{-1}MV$ has one of the following forms:

$$J = \begin{bmatrix} \lambda_1 & M_{12} & 0 \\ 0 & \lambda_2 & M_{23} \\ 0 & 0 & \lambda_3\end{bmatrix} \ \ \ \ \ \ \ \ J = \begin{bmatrix} \lambda_1 & 0 & 0 \\ 0 & a & b \\ 0 & -b & a\end{bmatrix},$$

where all the entries are real, $M_{12}, M_{23} \in \{0,1\}$, $\lambda_i \neq 0$ and $b \neq 0$.

In the second case, since all eigenvalues are distinct ($b \neq 0$), $S$ is diagonalizable over $\mathbb{C}$. We can take $S= V J V^{-1}$, $U= I$ and we are finished.

Now for the first case. Since this is the Jordan form, we have $(\lambda_1 \neq \lambda_2) \Rightarrow M_{12} = 0$, and similarly $(\lambda_2 \neq \lambda_3) \Rightarrow M_{23} = 0$.

Let $\Sigma = \mathbb{diag}(\lambda_1, \lambda_2, \lambda_3)$, and let $W = I + \alpha E_{12} + \beta E_{23}$, where $E_{12}$ is the matrix with zeros everywhere except a one in the $1,2$ position, and similarly for $E_{23}$. We want to determine constants $\alpha, \beta$ so that $J = \Sigma W = W \Sigma$. We have: $$ \Sigma W = \Sigma + \alpha \lambda_1 E_{12} + \beta \lambda_2 E_{23}\\ W \Sigma = \Sigma + \alpha \lambda_2 E_{12} + \beta \lambda_3 E_{23}$$

This shows that $\Sigma W = W \Sigma$ iff $\alpha( \lambda_1 - \lambda_2) = 0$ and $\beta( \lambda_2 - \lambda_3) = 0$. Furthermore, we want J$ = \Sigma W$, and this will be true iff $M_{12} = \alpha \lambda_1$, and $M_{23} = \beta \lambda_2$.

Let $\alpha = \frac{M_{12}}{\lambda_1}$ and $\beta = \frac{M_{23}}{\lambda_2}$, then it is easy to verify that all the conditions are satisfied. Then we can take $S = V \Sigma V^{-1}$, and $U = V W V^{-1}$, and we are finished.

share|cite|improve this answer
excuse me for my low level, but where can I find information related to the real normal form you are talking about? Is it just the rational canonical form? – Bombyx mori Aug 2 '12 at 6:57
It is real Jordan form, unfortunately my library is off the shelves at the moment so I can't give you a book reference, but wikipedia has one It follows from the usual (possibly complex) Jordan form. – copper.hat Aug 2 '12 at 7:00

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.