We are given $A,B \in Mat_n(\mathbb R)$ are symmetric positive-definite matrices such that $AB=BA$

Show that $AB$ is positive-definite

What I did:

First I showed that $AB$ is symmetric, this is easily shown from $(AB)^t=B^tA^t=BA=AB$

Now I'm trying to think, why are the eigenvalues of $AB$ all positive? Just because the eigenvalues of $A$ and $B$ are positive does not imply that $AB$'s eigenvalues are.

However, from sylvester's criterion, we know that all the leading principal minors of $A$ and $B$ have a positive determinant, and since the determinant of the product is the product of the determinants, we can infer that every leading principal minor of $AB$ is positive. Proof:

$det(AB_{ii})=det(A_{ii})det(B_{ii}) \geq 0$ since $\forall i, det(A_{ii}),det(B_{ii}) \geq 0$

So $AB$ is a symmetric matrix where every leading principal minor has a positive determinant, and so $AB$ is indeed positive-definite.

This is kind of a roundabout way of solving it, is there a way of actually showing the eigenvalues of $AB$ are positive? Did my solution even make sense?

  • $\begingroup$ How did you get that they are symmetric am I missing something? Edit: Just re-read the title ;-) $\endgroup$ – Ben Feb 4 '14 at 15:04
  • $\begingroup$ Ah yeah I forgot to write that, thanks for mentioning, $A$ and $B$ are symmetric. $\endgroup$ – Oria Gruber Feb 4 '14 at 15:04

$AB=BA$ is a interesting condition. Let $v$ be an eigenvector of $A$ such that $Av=\lambda v$. Then $A(Bv)=B(Av)=B(\lambda v)=\lambda(Bv)$. This implies that the eigenvector space $V_{\lambda}$ of $A$ is invariant space of $B$. Now we can find all eigenvalues of $B$ on $V_{\lambda}$. Take the basis of $V_{\lambda}'s$ for both $A$ and $B$, you will find what you need.

  • $\begingroup$ Since $A$ and $B$ are diagonlizable and $AB=BA$, there is a basis that diagonlizes both of them, so there is a $P$ such that $A=PD_AP^{-1}$ and $B=PD_BP^{-1}$, and so $AB=PD_AD_BP^{-1}$, and $D_AD_B$ is a diagonal matrix with only positive values on the diagonal, and so $AB$ is positive definite. $\endgroup$ – Oria Gruber Feb 4 '14 at 15:33
  • $\begingroup$ Yes, you are right. The above explaination is just to show how to diagonalize $A, B$ at the same time. $\endgroup$ – Wei Zhou Feb 4 '14 at 15:37

It is not true that $\det(AB)_{ii}=\det A_{ii}\det B_{ii}$ as the following example shows $A=B=\begin{pmatrix}2&1\\1&1\end{pmatrix}$, so $AB=\begin{pmatrix}5&3\\3&2\end{pmatrix}$ and $\det(AB)_{11}=5\neq 2\cdot2=\det(A_{11})\det(B_{11})$

So your argument has a gap.

Try to use the Spectral Theorem.

  • $\begingroup$ Yeah that's true...Damnit. $\endgroup$ – Oria Gruber Feb 4 '14 at 15:23

More generally, if $A,B$ are any real $n\times n$ matrices s.t. $AB=BA$ and $A,B$ have only positive eigenvalues, then $A+B$ and $AB$ have positive eigenvalues.

Proof: The $2$ conditions imply that $A,B$ are simultaneously triangularizable over $\mathbb{R}$. Then we may assume that they are upper-triangular with positive diagonals and we are done.


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.