During my study leave in 2018 I have applied nonlinear stability analysis techniques to the Douglas-Rachford Algorithm, with the aim of shedding light on the interesting non-convex case, where convergence is often observed but seldom proven. The Douglas-Rachford Algorithm can solve optimisation and feasibility problems, provably converges weakly to solutions in the convex case, and constitutes a practical heuristic in non-convex cases. Lyapunov functions are stability certificates for difference inclusions in nonlinear stability analysis. Some other recent nonlinear stability results are showcased as well.

We consider monotone systems defined by ODEs on the positive orthant in $\mathbb{R}^n$. These systems appear in various areas of application, and we will discuss in some level of detail one of these applications related to large-scale systems stability analysis.

Lyapunov functions are frequently used in stability analysis of dynamical systems. For monotone systems so called sum- and max-separable Lyapunov functions have proven very successful. One can be written as a sum, the other as a maximum of functions of scalar arguments.

We will discuss several constructive existence results for both types of Lyapunov function. To some degree, these functions can be associated with left- and right eigenvectors of an appropriate mapping. However, and perhaps surprisingly, examples will demonstrate that stable systems may admit only one or even neither type of separable Lyapunov function.