Multiple steady state solutions of a circular neural network (Q937345)

The main results are concerned with the following discrete time system, which generalizes the equations of some discrete neural networks with circular structure \[ X_{n+1} + X_n - f_n(X_n) = 0,\quad n\in Z \] with \(X\) and \(f\) \(k\)-vectors, \(\{f_n(X)\}_n\) being a \(\omega\)-periodic sequence. Assume there exists \(\{F_n(U)\}_n\) such that \(\nabla_UF_n(U)=f_n(U)\), \(\forall U\) and \(\{F_n(U)\}_n\) is also \(\omega\)-periodic. Assume also that \(F_n(U)\geq 0\), \(\forall n\in Z\), \(U\in R^k\) and i) there exist \(\delta>0\), \(\alpha\in(0,1)\) such that \(\forall n\in Z\), \(| U| \leq\delta\) then \(F_n(U)\leq\alpha| U| ^2\); ii) there exist \(\rho>0\), \(\gamma>0\), \(\beta\in(1,\infty)\) such that \(\forall n\in Z\), \(| U| \geq\rho\) then \(F_n(U)\geq\beta| U| ^2-\gamma\). Then the system has at least \(2k\omega\) non-trivial \(\omega\) solutions.