I was asked the following question:

Let $A,B \in \mbox{Mat}_n(\mathbb R)$ be symmetric positive definite matrices. Show that $AB$ does not have negative eigenvalues.

My answer

I'm getting something very weird. According to my answer, $AB$ has to be positive definite. which is weird because we don't know that $AB$ is symmetric.

$B$ is symmetric so there is a basis $v_1,...,v_n$ of eigenvectors such that $Bv_i=\lambda_i$ and $\lambda_i>0$ for all $i$, and $<v_i,v_j>=0$ when $i\neq j$

let's look at $<ABx,x>$ where $x$ is some vector in $\mathbb R^n$

$<ABx,x>=<AB\sum_{i=1}^n \alpha_i v_i,\sum_{i=1}^n \alpha_i v_i> = <AB\alpha_1 v_1,\alpha_1 v_1>+<AB\alpha_2 v_2,\alpha_2 v_2>+...+<AB\alpha_n v_n,\alpha_n v_n> = \alpha_1 ^2 \lambda_1<Av_1,v_1>+\alpha_2 ^2\lambda_2<Av_2,v_2>+...+\alpha_n ^2 \lambda_n<Av_n,v_n>$

obviously for all $i$ $\alpha_i ^2$ is positive.

Because $B$ is positive definite, for all $i : \lambda_i >0$.

Because $A$ is positive definite, for all $i: <Av_i,v_i>$ is positive.

so overall, for all $i: \alpha_i ^2 \lambda_i <Av_i,v_i>$ is positive.

So we got that $<ABx,x>$ is larger than zero for any vector $x$, and so I infer that $AB$ is positive definite. And so it doesn't have negative eigenvalues.

But how can we say that it's positive definite if it's not symmetric? I did something wrong I'm sure.


1 Answer 1


The matrix $AB$ is similar to a symmetric positive definite matrix $A^{-1/2}(AB)A^{1/2}=A^{1/2}BA^{1/2}$. So, its eigenvalues are real and positive.

  • 1
    $\begingroup$ Just in case the OP doesn't know how to find the square root of a matrix, I would like to add that if $U$ is an invertible matrix such that $UAU^{-1} = D$ is the diagonalization of $A$, we write $A^{1/2} = U^{-1} D^{1/2} U$ where $D^{1/2}$ is the diagonal matrix of the square roots of the diagonal entries of $D$. $\endgroup$
    – Joel
    Feb 10, 2014 at 14:49

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.