i'm a bit confused when trying to calculating generalized eigenvectors if there is more than one eigenvectors for a specific eigenvalue.

Example: $A=\begin{pmatrix}0&1&-1&-1\\0&0&0&0\\0&-1&2&2\\0&1&-2&-2\end{pmatrix}$

This matrix has the only eigenvalue $\lambda=0$ with algebraic multiplicity $m=4$


so geometric multiplicity $g=2$

From what I know, I'd say, that we now should be able to find two generalized eigenvectors. ($4-2=2$)

What I usually do to calculate generalized eigenvectors, if we have an eigenvector $x_1$ to some eigenvalue $p$ is:

$(A-pI)x_1=0$ [gives us the ordinary eigenvector]



so that we get the generalized eigenvectors $x_2,x_3$

Back to my example: If I do this: (Note that $(A-\lambda I)=A$

$Aw_1=v_1$ gives:

$w_1\in \{\begin{pmatrix}0\\1\\1\\0\end{pmatrix}+t\begin{pmatrix}1\\0\\0\\0\end{pmatrix}+s\begin{pmatrix}0\\0\\-1\\1\end{pmatrix} | s,t\in\mathbb R\}$

$Aw_2=v_2$ gives:

$w_2\in \{\begin{pmatrix}0\\2\\1\\0\end{pmatrix}+o\begin{pmatrix}1\\0\\0\\0\end{pmatrix}+u\begin{pmatrix}0\\0\\-1\\1\end{pmatrix} | u,o\in\mathbb R\}$

We choose $t=s=o=u=1$ and get

$w_1 = \begin{pmatrix}1\\1\\0\\1\end{pmatrix}$

$w_2 = \begin{pmatrix}1\\2\\0\\1\end{pmatrix}$

So, we get the matrix$ M=(v_1,w_1,v_2,w_2)$

Is that correct? How do I go about such a case?


The eigenspace for the eigenvalue $\lambda=0$ is given by: $$ A\mathbf x= \begin{bmatrix}0&1&-1&-1\\ 0&0&0&0\\ 0&-1&2&2\\ 0&1&-2&-2 \end{bmatrix} \begin{bmatrix} x\\y\\z\\t \end{bmatrix}= \begin{bmatrix} 0\\0\\0\\0 \end{bmatrix} $$ that gives: $$ \begin{bmatrix} x\\y\\z\\t \end{bmatrix}= \begin{bmatrix} x\\0\\-t\\t \end{bmatrix} $$ so we can chose two linearly independent eigenvectors as: $$\mathbf v_1= \begin{bmatrix} 0\\0\\-1\\1 \end{bmatrix}\qquad \mathbf v_2= \begin{bmatrix} 1\\0\\0\\0 \end{bmatrix} $$ Now using $\mathbf v_1$ we can find a generalized eigenvector searching a solution of: $$ \begin{bmatrix}0&1&-1&-1\\ 0&0&0&0\\ 0&-1&2&2\\ 0&1&-2&-2 \end{bmatrix} \begin{bmatrix} x\\y\\z\\t \end{bmatrix}=\begin{bmatrix} 0\\0\\-1\\1 \end{bmatrix} $$ that gives a vector of the form $$ \begin{bmatrix} x\\y\\z\\t \end{bmatrix}= \begin{bmatrix} x\\-1\\-1-t\\t \end{bmatrix} $$ and, for $x=t=0$ we can chose the vector $\mathbf w_1=[0,-1,-1,0]^T$

In the same way we can find the generalized eigenvector $\mathbf w_2=[0,2,1,0]$ as a solution of $A\mathbf x=\mathbf v_2$.

Now we have the matrix $$ M=[\mathbf v_1,\mathbf w_1,\mathbf v_2, \mathbf w_2]= \begin{bmatrix} 0&0&1&0\\ 0&-1&0&2\\ -1&-1&0&1\\ 1&0&0&0 \end{bmatrix} $$ with the inverse: $$ M^{-1}=\begin{bmatrix} 0&0&0&1\\ 0&1&-2&-2\\ 1&0&0&0\\ 0&1&-1&-1 \end{bmatrix} $$ and a Jordan decomposition of The matrix $A$ is: $$A= \begin{bmatrix}0&1&-1&-1\\ 0&0&0&0\\ 0&-1&2&2\\ 0&1&-2&-2 \end{bmatrix}= \begin{bmatrix} 0&0&1&0\\ 0&-1&0&2\\ -1&-1&0&1\\ 1&0&0&0 \end{bmatrix} \begin{bmatrix} 0&1&0&0\\ 0&0&0&0\\ 0&0&0&1\\ 0&0&0&0 \end{bmatrix} \begin{bmatrix} 0&0&0&1\\ 0&1&-2&-2\\ 1&0&0&0\\ 0&1&-1&-1 \end{bmatrix}= MJM^{-1} $$

This decomposition is not unique, In the sense that the matrix $M$ ( and $M^{-1}$) can be different, because we can chose different eigenvectors and generalized eigenvectors.

If, as in OP, we chose: $$\mathbf v'_1= \begin{bmatrix} 1\\0\\0\\0 \end{bmatrix}\qquad \mathbf v'_2= \begin{bmatrix} 1\\0\\1\\-2 \end{bmatrix} $$ than the generalized eigenvectors that satisfies the equations: $$ A\mathbf w'_1=\mathbf v'_1 \qquad A\mathbf w'_2=\mathbf v'_2 $$ becomes: $$ \mathbf w'_1= \begin{bmatrix} 1\\2\\0\\1 \end{bmatrix} \qquad \mathbf w'_2= \begin{bmatrix} 1\\3\\1\\1 \end{bmatrix} $$

(this seems the mistake in OP) and we have a matrix $$ S= \begin{bmatrix} 1&1&1&1\\ 0&2&0&3\\ 0&0&1&1\\ 0&1&-1&1 \end{bmatrix} $$ and a Jordan decomposition: $$ SJS^{-1}=\begin{bmatrix} 1&1&1&1\\ 0&2&0&3\\ 0&0&1&1\\ 0&1&-1&1 \end{bmatrix} \begin{bmatrix} 0&1&0&0\\ 0&0&0&0\\ 0&0&0&1\\ 0&0&0&0 \end{bmatrix} \begin{bmatrix} 1&-2&2&3\\ 0&2&-3&-3\\ 0&1&-1&-2\\ 0&-1&2&2 \end{bmatrix}= \begin{bmatrix}0&1&-1&-1\\ 0&0&0&0\\ 0&-1&2&2\\ 0&1&-2&-2 \end{bmatrix}=A $$

Finally, note that the eigenspace of the eigenvalue $\lambda=0$ is the kernel of $A$ and also $\mathbf u_1=A\mathbf e_2$ and $\mathbf u_2=A\mathbf e_3$ are vectors of the kernel, so they are eigenvectors of $A$, and $\mathbf e_2$ and $\mathbf e_3$ are the corresponding generalized eigenvectors, so another matrix that gives a Jordan decomposition is $N=[\mathbf u_1,\mathbf e_2,\mathbf u_2,\mathbf e_3]$

  • $\begingroup$ so the process itself was right? $\endgroup$ – xotix Feb 4 '17 at 21:10
  • $\begingroup$ Yes, the process is correct. See: en.wikipedia.org/wiki/Generalized_eigenvector $\endgroup$ – Emilio Novati Feb 5 '17 at 9:38
  • $\begingroup$ hmm, I'm still confused. What was my mistake? You have $v_1=(0,0,-1,1)$ and you have $v_1=(0,0,1,-1)$ but both are in the kernel, so it shouldn't matter, should it? $\endgroup$ – xotix Feb 5 '17 at 10:10
  • $\begingroup$ Also: The solution of me does this: We say $A^2=0$. Then they complement $kern(A-\lambda I)$ with the vectors $e_2, e_3$ to aBasis of $\mathbb R^4$. Note that $e_2,e_3\not\in Kern(A-\lambda I)$. But then $Ae_2$ and $Ae_3$ are in $Kern(A-\lambda I)$ So they get $S=(Ae_2,e_2,Ae_3,e_3)$. What exactly happens here? $\endgroup$ – xotix Feb 5 '17 at 10:13
  • $\begingroup$ Right! The mistake is not in the first step. I've edit my answer and added something. $\endgroup$ – Emilio Novati Feb 6 '17 at 10:42

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.