Consider the matrix $A = \begin{pmatrix} 2 & -3 \\ 1 & 4\end{pmatrix}$, which should represent a system of linear differential equations. I need to find the flow $\varphi(t, X)$ whereas $X$ is the initial value. Thus I need to calculate the exponential matrix $e^{tA}$.

In order to do that I determined the eigenvalues $\lambda_{1,2} = 3 \pm i \sqrt{2}$. Here is where I encounter some problems. I think I could determine the exponential matrix by saying: $A = SDS^{-1}$, whereas $D$ is the corresponding rotation matrix and $S$ consists of the (complex) eigenvectors of $A$ (and than using some characteristics of the exponential matrix).

However, I found out that apparently one can write $A$ as $A = D + N$ where $D$ is diagonalizable and $N$ is nilpotent (and $DN = ND$), since the characterstic polynom of $A$ can be written as linear factors. This approach seems to be easier than the solution I have noted above - depending on how easy it is to determine $D$ and $N$. Thus, my question: How do I determine $D$ and $N$? Is there a standard method or is it just "seeing"/guessing how you could obtain those two matrices?

Thanks in advance for your help!

  • $\begingroup$ Does the decomposition $A=D+N$ have a name? $\endgroup$ – Git Gud Feb 26 '13 at 20:39
  • $\begingroup$ @GitGud I'm not sure. On Wikipedia I found that if the characteristic polynom can be written as linear factors (which is always the case in $\mathbb C$), than such a decomposition exists. They didn't mention a name though. $\endgroup$ – kafman Feb 26 '13 at 20:57
  • $\begingroup$ How can I find it. Are you wanting to finding this decomposition or would an alternative be good as well? $\endgroup$ – Git Gud Feb 26 '13 at 20:59
  • $\begingroup$ @GitGud See here, point "Generalization". Actually, I was wondering how to find this decomposition. But, of course if you know of an alternative (which is not the one I've already mentioned), I'd be grateful as well. $\endgroup$ – kafman Feb 26 '13 at 21:05
  • 1
    $\begingroup$ I don't think there's a quick answer to what you're looking for. You probably gotta go with diagonalizability. Even JNF wouldn't save you from complex eigenvalues. I don't think there's a way around that. $\endgroup$ – Git Gud Feb 26 '13 at 21:16

Note that since $N$ is nilpotent, its only eigenvalue is $0$. If you were to write $A$ as $A=D+N$, you'd have $D=A-N$ which would imply that the eigenvalues of $D$ are those of $A$. So when you diagonalize $D$ you'll get complex eigenvalues again, so might as well use $A$'s diagonalizability from the get go. I don't think there's a way around the complex eigenvalues.


Since you already know that alleigenvalues of $A$ are distinct (i.e. there is no multiple eigenvalue, the matrix $A$ itself is diagonalizable, i.e. $D=A, N=0$. Go ahead with your first ideas, actually diagonalizing $A$ (see? It's diagonalizable).

  • $\begingroup$ "Thus, my question: How do I determine $D$ and $N$?" $\endgroup$ – Git Gud Feb 26 '13 at 20:15
  • $\begingroup$ But $A$ has complex eigenvalues, so I think $A$ is not diagonalizable, at least not in $\mathbb R$, right? Thus I couldn't say $D = A$. I was actually just wondering how it is possible to decompose $A$ into $D$ and $N$ ... $\endgroup$ – kafman Feb 26 '13 at 21:01

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.