On a method of mathematical modeling of electrical synapses (Q2082797)

This paper introduces a new mathematical model of a neural network (of \(m\geq 2\) neurons) with electrical synapses. The model is a singularly perturbed system of delay differential equations of the form \[\dot{u}_j=\left[\lambda f\left(u_j(t-1)\right)+\sum_{\substack{s=1 \\ s \neq j}}^m d_{j, s} g\left(\frac{u_s}{u_j}\right)\right] u_j, \quad j=1, \ldots, m\] where \(u_j\) (\(j=1,\dots,m\)) represents the potential of the \(j\)th neuron, \(\lambda>0\) and \(d_{j,s}\in \mathbb{R}\) are constant parameters and the function \(g(u) \in C^3\left(\mathbb{R}_{+}\right)\) satisfies the conditions \[ \begin{gathered} g(u)<0 \quad \text {for } \quad u \in[0,1), \quad g(0)=-1, \quad g(u)>0 \quad \text {for } \quad u>1, \quad g(1)=0, \quad g^{\prime}(1)>0 \\ g(u)-b, u g^{\prime}(u), u^2 g^{\prime \prime}(u), u^3 g^{\prime \prime \prime}(u)=O(1 / u) \quad \text {as } \quad u \rightarrow+\infty, \quad b=\text {const.}>0 \end{gathered} \] Some methods for studying existence and stability of periodic solutions are presented. In particular, it is shown that for the case of diffusion and symmetric fully connected neural networks, i.e.\ when there exists \(d>0\) such that \(d_{j,s}=d\) for all \(j\neq s\), the well-known buffering phenomenon can be realized in the model. This means that for certain parameters with \(m\) increasing, one can guarantee the coexistence of any preassigned finite number of stable cycles in the model. This phenomenon is a desirable feature of neural network models as it reflects the situation in which various ideas and concepts compete with each other for dominance in the neocortex of the human brain.