I have two $n\times n$ Hurwitz stable matrices, so $A_i\in H_n$ for $i=1,2$. I also know that every possible convex combination of these two matrices is Hurwitz, i.e. \begin{equation} C(A_1,A_2) = cA_1+(1-c)A_2 \in H_n \hspace{0.5cm} \forall c\in[0,1]. \end{equation} I would like to determine tractable conditions under which a common quadratic Lyapunov function exists for the unforced systems $\dot{x}(t)=A_i x(t)$ such that \begin{equation} A_iP+PA_i^T = -Q_i \hspace{0.5cm} i=1,2 \end{equation} with $P=P^T>0$ and $Q_i=Q_i^T>0$.

I am familiar with some results that might help here, but so far nothing is applicable. The matrices do not commute. I cannot establish that they are simultaneously triangularizable. They are generated from nonminimal systems and it is necessary that they are not expressed in companion form. They are of order $n$ which will certainly be greater than 2. Perhaps this is not possible. However, I was hoping that I could exploit the known stability of all convex combinations somehow.

I also have a related formulation that might be more feasible. Say the problem is now framed as an unforced time-varying system $\dot{x}(t)=A(t)x(t)$ with \begin{equation} A(t) = c(t)A_1 +(1-c(t))A_2 \hspace{0.5cm} c(t)\in[0,1]. \end{equation} The eigenvalues of $A(t)$ are always in the open LHP and $c(t)$ behaves nice enough to ensure that the system is exponentially stable. I would like to determine conditions under which a constant Lyapunov matrix exists \begin{equation} A(t)P+PA^T(t) = -Q(t). \end{equation} I view this as somewhat akin to finding a common Lyapunov function for a switched system that can switch between an infinite number of modes. Maybe I could restrict the time variation of $c(t)$ to help make this possible?

Edit: Looking at this from the time-varying system perspective, I think a converse Lyapunov theorem could help. If the system is exponentially stable for all permissible $c(t)$ trajectories then can I claim there exists a Lyapunov function satisfying \begin{equation} c_1\|x\|^2 \le V(x,t) \le c_2 \|x\|^2 \end{equation} \begin{equation} \dot{V}(x,t) \le -c_3 \|x\|^2 \end{equation} with $c_1,c_2,c_3>0$? Could I further claim that the form of such a function is \begin{equation} V(x,t) = x^TP(t)x \end{equation} with a smooth $P(t)=P^T(t)>0$ for an appropriate $c(t)$? Would there be any loss of generality in assuming this $P(t)$ to be a specific $P(t)$, i.e. one satisfying properties of another result I wish to use?

Edit 2: Is there any tractable way to relate the state transition matrices for $A_1$, $A_2$, and $A(t)$? I believe that I would have to work directly from the $\dot{\Phi} = A(t)\Phi$ statement to determine its transition matrix which seems daunting.

  • $\begingroup$ I believe the fact that each convex combination is Hurwitz is only useful when they commute. The only thing I could think of to solve this problem would be LMI's. $\endgroup$ – Kwin van der Veen Sep 16 '17 at 4:22


R. Shorten and K. Narendra, “Necessary and sufficient conditions for the existence of a common quadratic Lyapunov function for two stable second order linear time-invariant systems,” in Proc. Amer. Control Conf. , 1999, pp. 1410–1414.


R. Shorten, K. Narendra, and O. Mason, “A result on common quadratic Lyapunov functions,”IEEE Trans. Automat. Control , vol. 48, no. 1, pp. 110–113, Jan. 2003

The results are summarized in Theorem 1 of

Lin, H. and Antsaklis, P.J. (2009). Stability and stabiliz-ability of switched linear systems: A survey of recent re-sults.IEEE Transactions on Automatic Control , 54(2), 308–322. doi:10.1109/TAC.2008.2012009.

  • $\begingroup$ Thanks for the references. I am familiar with these results, but have so far been unable to find anything applicable to my situation. The more I think about it the more I doubt that that I can do anything along the common Lyapunov function line of thought. Even in the 2nd order case, it has already been shown that convex combinations of $A_1$, $A_2$, $A_1^{-1}$, and $A_2^{-1}$ are necessary and sufficient for a CLF. I don't even have the ability to say that much. The time-varying representation seems to be the way to go unfortunately. $\endgroup$ – user480735 Sep 20 '17 at 20:33

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.