1
$\begingroup$

Let $A$ be square or rectangular and $x\in \mathbb{R}$. Can you point me to equations/areas out there where $(AA^T)^x$ or $(A^TA)^x$ or their eigenvalues are used as applications? e.g. we find them in the polar decomposition when $x=\frac{1}{2}$.

I'm particular about cases when $x\ne 0, \frac{1}{2}, 1$. Thanks in advance.

$A$ may also be taken to be complex i.e. $(AA^*)^x$ or $(A^*A)^x$.

$\endgroup$
  • $\begingroup$ what does $A^x$ mean for real $x$? $\endgroup$ – Pink Panther Mar 17 at 14:14
  • $\begingroup$ Would you be willing to use a formula in terms of the eigenvalues of $A^TA$? There's also a formula using a power series. $\endgroup$ – Omnomnomnom Mar 17 at 14:36
  • $\begingroup$ @Omnomnomnom. Yes, I will consider their eigenvalues too. I have edited the post to include that $A$ can also be rectangular. $\endgroup$ – Kay Mar 17 at 15:47
  • $\begingroup$ @Omnomnomnom, please expantiate on the formula using a power series, and that which considers the eigenvalues. Thanks in advance. $\endgroup$ – Kay Mar 17 at 18:01
2
$\begingroup$

I'll stick to the real case, but the situation for complex matrices is similar.

$A^TA$ is a symmetric and positive semidefinite matrix. By the spectral theorem, there exists an orthogonal $U$ such that $U^T(A^TA)U$ is diagonal, that is $$ U^T(A^TA)U = D = \pmatrix{\lambda_1 \\ & \ddots \\ && \lambda_n} $$ where the $\lambda_k$ are the eigenvalues of $A^TA$ (i.e. the singular values of $A$). Because $A^TA$ is positive semidefinite, the eigenvalues $\lambda_k$ are non-negative, that is $\lambda_k \geq 0$ for all $k$.

We may compute $$ (A^TA)^x = [U D U^T]^x = UD^x U^T = U \pmatrix{\lambda_1^x \\ & \ddots \\ && \lambda_n^x}U^T $$ Because $\lambda_k\geq 0$ for all $k$, $\lambda_k^x$ makes sense for all $x > 0$. If we also know that $A^TA$ is invertible (i.e. that $A$ has full column rank), then we have $\lambda_k > 0$ which means that $\lambda_k^x$ makes sense for all $x \in \Bbb R$.

$\endgroup$
  • $\begingroup$ thanks for the response! I actually meant 'applications' when I said areas/equations where those relations 'are found'. I have edited the post accordingly. I'm sorry about the ambiguity! $\endgroup$ – Kay Mar 17 at 19:21

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.