Are the eigenvalues of the symmetric part of a matrix with eigenvalues – all with positive real part- positive too

eigenvalues-eigenvectorslinear algebramatricessymmetric matrices

Consider the invertible matrix $A$ with all eigenvalues $\lambda_{i}$ with positive real part. Then consider its symmetric part:
$$
A_{s}=\frac{A+A^{T}}{2},
$$

with real eigenvalues $\eta_{i}$. Are these eigenvalues $\eta_{i}$ all positive?

P.S. What I already know is that $\text{Trace}[A_{s}]=\text{Trace}[A]$.

Best Answer

No. E.g. the eigenvalues of $\dfrac12\left(\pmatrix{1&4\\ 0&1}+\pmatrix{1&0\\ 4&1}\right)=\pmatrix{1&2\\ 2&1}$ are $3$ and $-1$.

The converse is true, however. That is, if $A$ is a real square matrix such that $\frac{A+A^T}{2}$ is positive definite, then all eigenvalues of $A$ have positive real parts. To prove this, note that if $u$ is a unit eigenvector of $A$ corresponding to an eigenvalue $\lambda$, then $$ \Re(\lambda) =\frac{u^\ast(Au)+(u^\ast A^\ast)u}{2} =u^\ast\frac{A+A^\ast}{2}u =u^\ast\frac{A+A^T}{2}u>0. $$