I would like to show the following problem from linear algebra:

Let $n\in\mathbb{N}$ and let $A,B\in M_n$ be Hermitian matrices. If all eigenvalues of $A$ and $B$ are positive and $A^k=B^k$ for some $k\in\mathbb{N}$, then $A=B$.

I could understand that if $C$ is a Hermitian matrix, then eigenvalues of $C$ are real, and if $k=1$, then the statement is clear.

But I could not go further. I appreciate any advice. Thank you in advance.

  • $\begingroup$ Hint. You can define the $k$-th root of any positive semi-definite Hermitian matrices. $\endgroup$ – Sangchul Lee Jul 29 '17 at 10:59

We have

$$\ker(A-\lambda I_n)\subset \ker (A^k-\lambda^k I_n)$$ and since $A$ is Hermitian then diagonalizable so $$n=\dim\Bbb C^n=\sum_{\lambda\in\operatorname{Sp}(A)}\dim\ker(A-\lambda I_n)$$ Moreover, since $\lambda^k$ are distinct then $\sum\ker(A^k-\lambda^k I_n$) is a direct sum and so from the first inclusion we see that in fact $$\ker(A-\lambda I_n)=\ker(A^k-\lambda^k I_n)$$ The same reasoning for $B$ gives us finally $$\ker(A-\lambda I_n)=\ker(B-\lambda I_n)$$ and since $A$ and $B$ are diagonalizable we get $A=B$.

  • 1
    $\begingroup$ Thank you for the answer! But where did you use the hypothesis that eigenvalues are positive? $\endgroup$ – Tom TJ Jul 29 '17 at 11:47
  • 1
    $\begingroup$ The eigenvalues are positive then $\lambda^k$ are distinct whatever the value of $k$. $\endgroup$ – user296113 Jul 29 '17 at 13:28

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.