0
$\begingroup$

Suppose I have a matrix:

$A \in \mathbb{R}^{n \times n}$, where $A$ not necessarily symmetric, but $x^TAx \geq 0$, or $\leq 0$ for $x \in \mathbb{R}^n$.

Will such matrix have imaginary eigenvalues?

It seems that symmetry is a sufficient but not necessary condition for having real imaginary values. Will positive + negative (semi)definite plus being real ensure that $A$ will not have imaginary eigenvalues?

  • 0
    real imaginary values? real eigenvalues? – 2017-01-14

2 Answers 2

2

They can have imaginary eigenvalues. Consider, $$A = \left[ \begin{matrix} 1 & -1 \\ 1 & 1\end{matrix} \right].$$ For $x = \binom u v \in \mathbb R^2$, we see $$x^tAx = u^2 + v^2 \ge 0 $$ with equality iff $x = 0$. Thus $A$ is positive definite. However, the eigenvalue of $A$ are given by $$(\lambda -1)^2 + 1 = 0 \,\,\,\,\, \implies \,\,\,\,\, \lambda_{1,2} = 1 \pm i.$$ The real part of the eigenvalues of such a matrix will always be positive (resp. non-negative, negative or non-positive if the matrix is positive semi-definite, negative definite or negative semi-definite).

  • 0
    More generally, $$\begin{pmatrix}p&q\\-q&p\end{pmatrix}$$ with $p>q>0$ ought to work. – 2017-01-14
  • 0
    Yes you're right. Good point! – 2017-01-14
  • 0
    OP and @J.M.isn'tamathematician: Are there special cases for non symmetric matrices where you can make assertions about the imaginary parts of the eigenvalues? Please have a look at my question [Eigenvalues of the sum of two negative definite matrices with real eigenvalues](https://math.stackexchange.com/questions/2189690/eigenvalues-of-the-sum-of-two-negative-definite-matrices-with-real-eigenvalues). – 2017-03-20
0

To expand on User8128s answer, $\det(A-\lambda I)$ is a polynomial of real coefficients. Such a polynomial always has roots which are either real or come in complex conjugate pairs.