Suppose $A$ is a $n\times n$ real matrix, and the characteristic polynomial is $p(\lambda)=\det(A-\lambda I_n).$
By the Cayley-Hamilton theorem, $p(A)=0$. Here is my question:

Assume that $A=S\Lambda S^{-1}$, where $\Lambda=diag(\lambda_1,\cdots,\lambda_n)$, how can I get $p(A)=0$ by substitution?

I wrote $p(\lambda)=\prod_{i=1}^{n}(\lambda-\lambda_i)$. In the spirit of Polya, I try the most simply case.
When $n=2$, we have: $$p(A)=S\Lambda^2S^{-1}-(\lambda_1+\lambda_2)A+\lambda_1\lambda_2I_2.$$

But I have no idea how to go on.

Source: 6.2.32, P311, Intro to Linear Algebra 4th ed, by Gilbert Strang

  • 2
    $\begingroup$ Do you understand how to add and multiply diagonal matrices? This is an easier problem than you're making it. $\endgroup$ – Qiaochu Yuan Aug 5 '11 at 2:48
  • $\begingroup$ @Qiaochu: I guess you are talking about using $p(\lambda)=\sum_{k}^na_k\lambda^k$ rather than splitting it as indicated in anon's answer, aren't you? $\endgroup$ – Jack Aug 5 '11 at 3:01
  • 1
    $\begingroup$ I'm not sure what you mean by that. What I'm asking is whether you can see why $p(A) = \text{diag}(p(\lambda_1), ... p(\lambda_n))$. $\endgroup$ – Qiaochu Yuan Aug 5 '11 at 3:08

It's actually best, in my opinion at least, to use the usual polynomial expansion for $p$. Note that taking powers with a diagonalizable matrix is easy: $A^k = S\Lambda^k S^{-1}$ (by induction). Thus you can write $p(A) = \sum_{k=0}^n a_k A^k = \sum_{k=0}^n a_k S \Lambda^k S^{-1}=S(\sum_{k=0}^n a_k\Lambda^k)S^{-1}=S\,p(\Lambda)S^{-1}$. Finally, diagonal matrices in polynomials are easy: $p(\Lambda)=\mathrm{diag}(p(\lambda_1),p(\lambda_2),\dots,p(\lambda_n))$. But the eigenvalues of $A$ are preciesly the zeroes of $p(\cdot)$, so $p(\Lambda)=0$, and hence $p(A)=0$.

EDIT: If the problem specifically indicates for you to substitute the diagonalization in to the product, then that's what you should do. The idea behind that route is that the $(A-\lambda_iI)$'s can be refactored as $S(\Lambda-\lambda_i)S^{-1}$, making $p(A)=S(\Lambda-\lambda_1)(\Lambda-\lambda_2)\cdots(\Lambda-\lambda_n)S^{-1}$. Ignoring the outside factors, we know that the product of diagonal matrices is just the pairwise scalar product of their diagonal entries. Thus, using the product expression for $p(\cdot)$ in each of the diagonal entries, we find that $p(A)$ comes to $S\, \mathrm{diag}(p(\lambda_1),\cdots,p(\lambda_n))S^{-1}$, which is of course just $0$.

  • $\begingroup$ +1. I think so. It's somewhat "misleading" that the original problem which is from is "Substitute $A = S\Lambda S^{-1}$ into the product $$(A-\lambda_1 I)(A-\lambda_2I)\cdots(A-\lambda_nI)$$ and explain why this produces the zero matrix." $\endgroup$ – Jack Aug 5 '11 at 3:06
  • 2
    $\begingroup$ @Jack: In the OP you never mentioned the fact that the problem specifically asked for the substitution to be done inside the product form. That changes my answer, please see it now. $\endgroup$ – anon Aug 5 '11 at 3:17

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.