0
$\begingroup$

I'm trying to find the Jordan Normal Form of the matrix $A = \begin{bmatrix} 1 & 0 & 1 & -1 \\0 & 1 & 1 & 0\\ 0 & 0 & 1 & 1\\ 0 & 0 & 0 & 1\end{bmatrix}$

(That is, a matrix J and a matrix S such that $A = S J S^{-1})$.

Finding the eigenvalues and eigenvectors is no problem. The only eigenvalue is $\lambda$=1 with multiplicity 4. The corresponding eigenvectors are $x = \begin{bmatrix}1\\0\\0\\0\end{bmatrix}$ and $y = \begin{bmatrix}0\\1\\0\\0\end{bmatrix}$.

However, I get stuck after that. In order to find the S and J matrix, I have to determine the consistency of the matrices $(A-I)w = x$ and $(A-I)w = y$. But both of these matrices are inconsistent:

$(A-I)w = x \rightarrow \begin{array}{cccc|c}0 & 0 & 1 & -1 &1\\0 & 0 & 1 & 0 & 0\\0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0\end{array} \rightarrow \begin{array}{cccc|c}0 & 0 & 1 & 0 &1\\0 & 0 & 1 & 0 &0\\0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0\end{array}$

Which is clearly inconsistent since w3 = 0 from row 2 and w3 = 1 from row 1.

Also, $(A-I)w = x \rightarrow \begin{array}{cccc|c}0 & 0 & 1 & -1 &0\\0 & 0 & 1 & 0 & 1\\0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0\end{array} \rightarrow \begin{array}{cccc|c}0 & 0 & 1 & 0 &0\\0 & 0 & 1 & 0 &1\\0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0\end{array}$

Which is also clearly inconsistent.

So right now I do not know how to construct either the S or J matrix, since at least one of the systems $(A-I)w = x$ or $(A-I)w = y$ should have had a solution (and then also a solution for $(A-I)^2$ and possibly $(A-I)^3$ if there's one 1x1 and one 3x3 block).

Can someone help me out on how to construct the S and J matrices? Thanks!

$\endgroup$
  • $\begingroup$ Note that $x+y$ is also an eigenvector, and $(A-I)w=x+y$ does have a solution. $\endgroup$ – Gerry Myerson Apr 10 '14 at 10:35
1
$\begingroup$

Since you have found a $2$-dimensional eigenspace, you know the image of $A-I$ only has dimension$~2$ (that is the rank of $A-I$). You also know $A-I$ is nilpotent, so $(A-I)^2$ must have smaller rank (and equivalently larger kernel), the question is whether it drops to $0$ or becomes$~1$. Simple calculation show that $(A-I)^2$ is nonzero (so has rank $1$) and in fact its image is spanned by $(1,1,0,0)$ (the image of the last standard basis vector). So you've got a Jordan block of size $3$, and another one with size $1$ is necessarily left.

Now an important point to notice is that the first basis vector for the Jordan block of size$~3$ is not just any eigenvector (although it must be one), but also a vector in the image of $(A-I)^2$. The latter means it must be $b_1=(1,1,0,0)$, up to a scalar. Now the last basis vector $b_3$ for this Jordan block can be any vector with $(A-I)^2\cdot b_3=b_1$, and we saw that $b_3=(0,0,0,1)$ has this property. Then $b_2=(A-I)\cdot b_3=(-1,0,1,0)$ is also clear. Finally $b_4$ is just any eigenvector linearly independent from $b_1$, so $b_4=(1,0,0,0)$ will do. This gives a Jordan basis.

$\endgroup$
  • $\begingroup$ Hi Marc. Great answer, I'd like to get the general steps of your method...could you perhaps expand on that a little? In determining the matrix $S$ specifically. For example should the eigenvector which is the first basisvector for a larger size Jordan block always be in the image of $(A-\lambda I)^2$... $\endgroup$ – Christiaan Hattingh Apr 10 '14 at 15:10
  • $\begingroup$ Hi Marc. Never mind, thanks I get it, $b_1$ must be in the image of $(A-I)^2$ so that it won't be 'killed' (be in the kernel), else the chain stops there...elegant answer - I +1ed $\endgroup$ – Christiaan Hattingh Apr 10 '14 at 16:25
1
$\begingroup$

Here is another way to determine $J$.

  1. Find the characteristic polynomial. Since the matrix A is upper triangular it is easy to see that $c_A(x)=(x-1)^4$.
  2. Find the minimum polynomial. Every linear factor of $c_A(x)$ must also be a factor of $m_A(x)$. So we must just find $n$ so that $(A-I)^n=0$. Turns out $n=3$.

Now the multiplicity of $\lambda=1$ in the minimum polynomial corresponds to the size of the largest simple Jordan block, and so we must have \begin{equation} J=\begin{bmatrix} 1&0&0&0 \\ 1&1&0&0\\ 0&1&1&0 \\ 0&0&0&1 \end{bmatrix}. \end{equation} So now you have $J$ but you still need $S$. We have $AS=SJ$; let $S=[S_1,S_2,S_3,S_4]$ where each $S_i$ is the $i$th column of $S$. Then we have \begin{equation} AS=[AS_1, AS_2,AS_3,AS_4]=SJ=[S_1+S_2,S_2+S_3,S_3,S_4],\end{equation}so equating both sides (starting with the last column): \begin{equation} AS_4=S_4 \Leftrightarrow (A-I)S_4=0,\end{equation}\begin{equation} AS_3=S_3 \Leftrightarrow (A-I)S_3=0,\end{equation}\begin{equation} AS_2=S_2+S_3 \Leftrightarrow (A-I)S_2=S_3 \Leftrightarrow (A-I)^2S_2=0,\end{equation}\begin{equation} AS_1=S_1+S_2 \Leftrightarrow (A-I)S_1=S_2\Leftrightarrow (A-I)^2S_1=S_3 \Leftrightarrow (A-I)^3S_1=0.\end{equation}

So $S_3$ and $S_4$ are two linearly independent eigenvectors of $A$ associated with $\lambda=1$.


From this point I must say thanks to Marc, who's elegant answer switched the light on for me (please if you +1 me, do so for him too). In my answer I just want to expand the general method to find $S$: so here goes...

  1. Since $S_1$ must be nonzero (The matrix $S$ must be invertible) and $(A-I)^2S_1=S_3$ we see that the eigenvector $S_3$ (which must also be nonzero) must be in the image of $(A-I)^2$. And we see that Ran$(A-I)^2=\{x(1,1,0,0)^T:x \in \mathbb{F}\}$. So let $S_3=(x_1,x_1,0,0)^T$ with $x_1 \neq 0$ (!).
  2. Now we can calculate $S_1$ using the equation in 1 above. We get the general solution $S_1=(x_2,x_3,x_4,x_1)^T$ (so the only restriction is that the last coordinate of $S_1$ must equal $x_1$, other parameters are free).
  3. Now we can calculate $S_2$ by $(A-I)S_1=S_2$. The general solution is $S_2=(x_4-x_1,x_4,x_1,0)^T$.
  4. Now all we need is that $S_4$ must be linearly independent from $S_3$. So let $S_4=(x_5,0,0,0)^T$.

You can check that \begin{equation} S=\begin{bmatrix} x_2 & x_4-x_1 & x_1 & x_5 \\ x_3 & x_4 & x_1 & 0 \\ x_4 & x_1 & 0 & 0 \\ x_1 & 0 & 0 & 0\end{bmatrix}, \end{equation} is such that $S^{-1}AS=J$

$\endgroup$

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.