1
$\begingroup$

Let $A\in \mathcal{M}_{m\times m}(\mathbb R)$ , $det(A)=1$ , $A$ is positively definite. Which matrices $P$ satisfy the equation $$P^TAP=A$$ In fact I am interested in sequences of traces $tr P^n$ of the iterations of such solutions.

In dimension $2$ one can show that $$P^n=\left( \begin{array}{cc} \cos n\phi& -\beta \sin n\phi \\ \alpha \sin n\phi & \cos n\phi\\ \end{array}\right)$$ for some $\phi, \alpha, \beta$ satisfying$\alpha\beta=1$ hence $tr P^n= 2\cos n\phi$.

$\endgroup$
1
  • $\begingroup$ Two comments: All invertible $P$ with this property is a group. The second comment:the expresion $P^TAP$ instead of $P^{-1}AP$ remind me of the following: after change of coordinate with linear part $P$ the first fundamental form of the surface would be replaced by $P^T\mathcal{F}P$ instead of $P^{-1}\mathcal{F}P$. $\endgroup$ Jun 23, 2021 at 21:30

2 Answers 2

3
$\begingroup$

As $A=L^\top L$, for some $L\in M_{m\times m}(\mathbb{R})$, $\det L=1$, you can rewrite $P^\top L^\top LP=L^\top L$ and then multiply both sides by $L^{-1}$, etc., obtaining $(L^{\top})^{-1}P^\top L^\top LPL^{-1}=(LPL^{-1})^\top LPL^{-1}=I$, i.e. each $LPL^{-1}$ must be orthogonal.

As traces are preserved under conjugation, you can assume $L=I$, and so your question is reduced to studying traces of orthogonal matrices. The trace is the sum of eigenvalues, which are all modulus 1 complex numbers for orthogonal matrices.

$\endgroup$
3
  • $\begingroup$ It would be interesting to compare the two lie groups: centralizer of A and the other one mentioned in the question. $\endgroup$ Jun 23, 2021 at 22:13
  • $\begingroup$ Centraliser of $A$, i.e. $\{X \in GL_m\mid AX=XA\}$ is a different beast - it depends on eigenvalues of $A$. It's quite small if they are all different, and the whole $GL_m$ is they are all the same. $\endgroup$ Jun 24, 2021 at 12:08
  • $\begingroup$ Yes I see. On the other hand the Lie algebra of the other group is $V^T A+AV=0$ $\endgroup$ Jun 24, 2021 at 14:32
2
$\begingroup$

Using the vec operator, any matrix $X$ such that $vec A$ is an eigenvector of $(X\otimes X)^T$ corresponding to the eigenvalue $1$ would work.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.