Dynamics of large cooperative pulsed-coupled networks (Q2438353)

The author considers networks \(\mathcal{N}\) of pulse-coupled cells. Each cell is described by a dynamical system and it has a \textit{satisfaction variable} \(S_i\) which, if not externally perturbed, is strictly increasing until it reaches a \textit{threshold value} \(\theta_i\). When this happens the value of \(S_i\) is reset to \(0\) and the cell produces a spike that induces a discontinuous change \(\Delta_{j,i} > 0\) to the satisfaction variable of each cell \(j \neq i\). Denote by \(t_n\) the instants when at least one cell spikes (spiking instants). Then \(\mathcal{N}\) \textit{eventually periodically synchronizes spikes} with period \(p \geq 1\) if there is a \(n_0\) such that (i) at \(t_{n_0 + hp}\) all cells spike together (ii) the spiking pattern repeats every \(p\) spiking instants. It is shown (Theorem 3.5) that if \(\mathcal{N}\) is large enough then the network eventually periodically synchronizes spikes and this behavior is robust. A \textit{code-pattern} \(\Pi_k\) of length \(k\) is a sequence of sets of spiking cells. The code-pattern \(\Pi_{n,k}\) is defined as \[ \Pi_{n,k} = (I_n,\dots,I_{n+k-1}), \] where \(I_n\) is the set of spiking cells at the spiking instant \(t_n\) and a code-pattern of length \(k\) is \textit{recurrent} if there exists a subsequence \(n_j\) such that \(\Pi_{n_j,k} = \Pi_k\) for all \(j\). The set of all recurrent code-patterns with length \(k\) obtained from all initial states of \(\mathcal{N}\) is denoted by \(\mathcal{P}_k\). Then the paper defines the \textit{amount of information that \(\mathcal{N}\) can process} by \[ H = \sup_{k\geq1} \{ \log_2(\#\mathcal{P}_k) \}. \] Theorem 3.20 proves the intuitively clear \(H = \log_2 p\). In the particular case \(p=1\) we get \(H=0\), expressing the fact that there is precisely one recurrent code-pattern. Subsequently, the paper defines two new quantities related to the death of a cell, that is, to the possibility that a continuing influence from outside the network may bring the cell to a state where it stops spiking. The first quantity, called the \textit{intrinsic risk of death} of a cell, is \(R_i = \theta_i / \Theta\) where \(\Theta = \max_{j\in\mathcal{N}} \theta_j\). The second, called the \textit{net risk of death} during an inter-spike interval, is defined as \(R_i' = [(\theta_i-\Delta_i)/\Theta]\) where \([x]\) is the value of \(x\) clamped between \(0\) and \(1\) and \(\Delta_i\) is the net action that the cell receives during the inter-spike interval. It follows immediately that \[ R_i' = [(1-\Delta_i/\theta_i) R_i], \] and for cooperative networks \(\Delta_i > 0\) giving \(R_i' < R_i\) (Theorem 3.15) and suggesting that the risk of death is decreased in cooperative networks.