A periodic solution with non-simple oscillation for an equation with state-dependent delay and strictly monotonic negative feedback (Q2178717)

This paper demonstrates that the Poincaré-Bendixson-type results of \textit{J. Mallet-Paret} and \textit{G. R. Sell} [J. Differ. Equations 125, No. 2, 441--489 (1996; Zbl 0849.34056)] do not carry over entirely to state-dependent delay equations. In more detail, let us consider equations of the form \[ x'(t)=f(x(t-d(x_t)),\tag{1} \] where \(x_t\) denotes a member of \(C([-1,0];\mathbb{R})\) given by \(x_t(s)=x(t+s)\), \(s\in [-1,0]\), and the delay function \(d\) is defined on an appropriate subset of \(C\). It is well-known that, if \(f\) is smooth and strictly monotonic, and the delay \(d\) is constant, then the Poincaré-Bendixson-type results of Mallet-Paret and Sell [loc. cit.] hold. That is, if \(x\colon \mathbb{R} \to \mathbb{R}\) is a bounded entire solution of \(x'(t)=f(x(t-1))\), then \(\omega(x_0)\) is either a single non-constant periodic orbit or a collection of equilibria and connections between them. Moreover, the map \(\Pi\colon \omega(x_0)\to \mathbb{R}^2\) given by \[\Pi(\phi)=(\phi(0),\phi(-1))\in\mathbb{R}^2\] is injective. In a companion paper [J. Differ. Equations 266, No. 4, 1865--1898 (2019; Zbl 1420.34088)], the author gives a subclass of equation (1) for which the analogue of the above described Poincaré-Bendixson-type results hold. The present paper demonstrates that these results do not carry over to all equations of the form (1) with smooth, monotonic right-hand side. More precisely, the following result is proved: Theorem. There is an instance of equation (1) for which the following hold. \begin{itemize} \item There is a subset \(X\subseteq C\) such that every \(x_0\in X\) has a unique continuation \(x\colon [-1,\infty)\to \mathbb{R}\) as a solution of equation (1), and \(x_t\in X\) for all \(t\geq 0\); \item \(f\) is continuously differentiable and strictly decreasing, with \(f(0)=0\); \item \(d\colon X\to (0,1]\) is Lipschitz continuous and, for every solution \(x\colon [-1,\infty)\to \mathbb{R}\) with \(x_t\in X\) for all \(t\geq 0\), \(t-d(x_t)\) is strictly increasing; \item There is a nontrivial periodic solution \(q\colon \mathbb{R} \to \mathbb{R}\), with \(q_t\in X\) for all \(t\in\mathbb{R}\), such that the map \(\Pi\colon \{q_\tau : \tau\in \mathbb{R}\}\to \mathbb{R}^2\) given by the formula \[\Pi(q_t)=(q(t),q(t-d(q_t)))\] is not injective. \end{itemize}