1: \begin{abstract}

2:   Physical systems experience nonlinear disturbances which have the potential

3:   to disrupt desired behavior.

4:   For a particular disturbance, whether or not the system recovers from the

5:   disturbance to a desired stable equilibrium point depends on system parameter

6:   values, which are typically uncertain and time-varying.

7: %  Therefore, it is valuable to quantify proximity to vulnerability in terms of

8: %  parameter values.

9:   Therefore, to quantify proximity to vulnerability

10:   we define the safety margin to be the smallest change in

11:   parameter values from a nominal value such that the system will no longer

12:   be able to recover from the disturbance.

13:   Safety margins are valuable

14:   but challenging to compute as related methods, such as those for robust

15:   region of attraction estimation,

16:   are often either overly conservative or computationally intractable for

17:   high dimensional systems.

18:   Recently, we developed algorithms to compute safety margins efficiently

19:   and non-conservatively by exploiting the large sensitivity of the system

20:   trajectory near the region of attraction boundary to small perturbations.

21:   Although these algorithms have enjoyed empirical success, they lack

22:   theoretical guarantees that would ensure their generalizability.

23:   This work develops a novel characterization of safety margins in terms of

24:   trajectory sensitivities, and uses this to

25:   derive well-posedness and convergence guarantees for these

26:   algorithms, enabling their generalizability and successful application

27:   to a large class of nonlinear systems.

28: \end{abstract}

29: