1
$\begingroup$

For example, if $A$ is positive definite there exists a square root matrix $A^\frac{1}{2}$ for which $A^\frac{1}{2}A^\frac{1}{2}=A$. Proof. Let $A$ be a positive definite matrix with positive eigenvalues. Using the spectral decomposition, we have $$A=U^Tdiag(\lambda_1,…,\lambda_n)U=(U^Tdiag(\sqrt{\lambda_1},…,\sqrt{\lambda_n})U)(U^Tdiag(\sqrt{\lambda_1},…,\sqrt{\lambda_n})U) =A^\frac{1}{2}A^\frac{1}{2} (\lambda_i>0)$$ original text

My Question is: Since the eigenvalues of a positive definite matrix are all positive, does a positive definite matrix have any power of real number, $i.e. $ does $$A^x, x\in R$$exist? Such as $A^\frac{1}{3}, A^{-\frac{1}{2}} or A^{-1}$?

Please give me a proof, rather than just examples.

$\endgroup$
5
  • $\begingroup$ Well, $A^{-1}$ is the inverse of $A$ and it exists when $\det\left(A\right)\neq 0$. $\endgroup$ Sep 10, 2015 at 7:20
  • $\begingroup$ Yes, it is. Because $A$ is positive definite and $det(A)=\prod_{i=1}^n\lambda_i>0$ $\endgroup$
    – BioCoder
    Sep 10, 2015 at 7:24
  • $\begingroup$ The answer to your question is yes. $\endgroup$
    – Qudit
    Sep 10, 2015 at 7:30
  • $\begingroup$ Also, I think we can generalize $a^z=e^{z \log\left( a \right)}$ for $a,z \in \mathbb C$ to something like $A^z = \exp \left( z \log \left( A \right) \right)$ with $\exp$ and $\log$ functions defined as power series of matrices, and choosing a branch of the complex logarithm. I think this would work for any invertible matrix. $\endgroup$ Sep 10, 2015 at 7:31
  • $\begingroup$ go to canonical jordan form and there you may raise to any power then go back. $\endgroup$
    – Adelafif
    Sep 10, 2015 at 7:37

1 Answer 1

2
$\begingroup$

If you proceed exacly in the same way as you found the square root replacing the $1/2$ by $s>0$, then you will obtain a matrix $U^T\operatorname{diag}(\dots,\lambda_i^s,\dots)U$ which deserves to be called the $s$th power of $A$.

Indeed, any sensible notion of $s$th power will be compatible with conjugation, so that $(C^{-1}AC)^s=C^{-1}A^sC$, and act in the obvious way on diagonal matrices.

$\endgroup$

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.