Consider a dynamic system $$\dot{x}=Ax+Bu \text{ and } y=Cx$$ The transfer function is $$sX(s) = AX(s)+BU(s),$$ so $$(sI-A)X(s)=BU(s)$$ and $$Y(s)=CX(s)$$ combining the two transfer equations, we have $$Y(s)=C(sI-A)^{-1}BU(s),$$ so $$G(s)=\frac{Y(s)}{U(s)}=C(sI-A)^{-1}B.$$
Now if, $G(s)$ has unstable zeros, which means zeros on the right half plane, section 1.3.4 "The Infinite Horizon LQ Problem" in book "Model Predictive Control: Theory, Computation, and Design 2nd Edition, Rawlings, et.al." said a LQR controller could invert this unstable zero to cause the system to become unstable. How is this possible? I know zeros of $G(s)$ is poles of $1/G(S)$, which is obvious. But how is this related to the system described above?
From the point of view of root-locus, infinite gain lead a pole to approach zero, but LQR always generate a finite value gain, what does the inversion of zeros mean in this context, could someone give some detailed explanation or maybe give some good reference?