0
$\begingroup$

The Jordan-Chevalley expresses a linear operator $M$ as $$ M = D + N, $$ where $D$ is semisimple (diagonalizable), $N$ is nilpotent and $DN=ND$. Although it is stated in many sources that $D$ and $N$ can be written as polynomials in $M$, I never see any method for obtaining such polynomials. I want to emphasize that I am not interested in a proof about the existence of these polynomials, but in a procedure for explicitly obtaining them given the Jordan-Chevalley decomposition.

$\endgroup$
  • $\begingroup$ If you already know $D$, $N$, expressing them as linear combination of $I,M,M^2,...,M^{n-1}$ is a simple Gaussian elimination problem. $\endgroup$ – user120527 Feb 13 at 10:56
  • $\begingroup$ That is not clear to me. Could you expand your comment into an explicit answer? $\endgroup$ – jobe Feb 13 at 11:04
1
$\begingroup$

Write $\chi_M(X)=\prod_{k=1}^s{(X-\lambda_k)^{\alpha_k}}$ (characteristic polynomial).

From among others Cayley-Hamilton, we know that $\bigoplus_{k=1}^s{\ker((M-\lambda_kI)^{\alpha_k})}=\mathbb{C}^n$ (for instance).

Let, for each $k$, $A_k$, $B_k$ be polynomials such that $A_k(X)(X-\lambda_k)^{\alpha_k}+B_k(X)\prod_{l \neq k}{(X-\lambda_l)^{\alpha_l}}=1$.

Let $P_k(X)=B_k(X)\prod_{l \neq k}{(X-\lambda_l)^{\alpha_l}}$.

Then, if $(M-\lambda_kI)^{\alpha_k}v=0$, then $P_k(M)v=v$. On the other hand, if $l \neq k$ and $(M-\lambda_lI)^{\alpha_l}v=0$, then $P_k(M)v=0$.

Then $D=\sum_k{\lambda_kP_k(M)}$.

$\endgroup$
  • $\begingroup$ I am having a hard time to fully understand your answer. One point: how can one compute $A_k(X)$ and $B_k(X)$? I have tried for some cases, but I could not. Is the relation between $A_k$ and $B_k$ a version of the Bézout's identity? $\endgroup$ – jobe Feb 13 at 20:42
  • $\begingroup$ Yes. That’s Euclid algorithm for polynomials. $\endgroup$ – Mindlack Feb 13 at 20:43
  • $\begingroup$ Ok. I will spend some additional time to completely understand your answer before accept it. Anyway, thank you very much! $\endgroup$ – jobe Feb 13 at 20:47

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.