Global asymptotic dynamics of a class of nonlinearly coupled neural networks with delays (Q379607)

The author considers a neural network consisting of two unidirectionally delay coupled rings, which are additionally coupled to each other via one component: \[ \begin{aligned}\dot{x}_{i}(t) &= -x_{i}(t)+g(x_{i-1}(t-\tau_{I})),i=1,\dots,K-1,\\ \dot{x}_{K}(t) &= -x_{K}(t)+g(x_{K-1}(t-\tau_{I}))+cg(y_{K}(t-\tau_{T})),\\ \dot{y}_{i}(t) &= -y_{i}(t)+g(y_{i-1}(t-\tau_{I})),i=1,\dots,K-1,\\ \dot{y}_{K}(t) &= -y_{K}(t)+g(y_{K-1}(t-\tau_{I}))+cg(x_{K}(t-\tau_{T})),\end{aligned} \] where \(g(x)\) is a bounded activation function from a rather general class of sigmoidal functions. The single components are scalar. The author derives several delay-dependent as well as delay independent criteria for the global synchronization and, more specifically, for the convergence towards in- and anti-phase equilibria or synchronous periodic solutions, respectively. These criteria are based on several asymptotic bounds for the single component's differences \(z_{i}=x_{i}-y_{i}\). In a second part, two four node networks with only one delay value, but inhomogeneous coupling strengths are considered. Also for these networks, global synchronization and possible attractors are studied and corresponding criteria are proven.