Metastable periodic patterns in singularly perturbed delayed equations (Q5961949)

The equation \[ \varepsilon\dot{x}(t)=-x(t)+f(x(t-1)) \] is considered in the limit \(\varepsilon\to 0\) for both cases of Positive Feedback (PF) and Negative Feedback (NF) by the nonlinearity \(f\), which is assumed to be odd in the positive feedback case. Metastable behavior in the following sense is described: Solutions with appropriate initial functions exhibit `pseudo-periodic' oscillations for times of order \(\mathrm{exp} (\mathrm{const}/\varepsilon)\) before they may eventually converge to an equilibrium. In both feedback cases, a central step is the construction of so-called transition layer solutions \(\phi\) and \(\chi\) which have the property that \(\phi_\varepsilon (t):=\phi(t/\varepsilon)\) and \(\chi_\varepsilon(t):=\chi(t/\varepsilon)\) provide good descriptions of the ascending/descending parts of typical oscillations. \(\phi\) and \(\chi\) are constructed in the paper for the PF case, while a result of Chow, Lin and Mallet-Paret can be used in the NP case, with modifications. Here are described shortly the main ideas of the results for the (pf) case (Theorem 2): A periodic approximate flow \(\varphi\) (with periods \(\approx 1\)) is constructed on a set \(W_n\) of initial functions with \(n\) zeroes; these `pseudo-solutions' are obtained by concatenating translates of \(\phi_\varepsilon\) and \(\chi_\varepsilon\). A projection map \(P_n\) from a set \(X_n\) of more general initial functions to \(W_n\) is defined via the zeroes such that \(P_n(x)\) has the same zeroes as \(x\). The difference between \(\varphi\) and the actual flow \(\psi\) for initial values in \(W_n\) over time approximately \(1\) is estimated. For a Poincaré-type map \(F_n:x\mapsto\psi_T(x)\), where \(T\) is the first time larger than one such that \(\psi_T(x)\in X_n\), the difference \(\|F_n(x)-(P_n\circ F_n)(x)\|\) is estimated (in both estimates by \(\exp(-\mathrm{const}/\varepsilon)\)). In addition, is is shown that one application of \(F_n\) changes the location of the zeroes of \(x\) only by an amount of order \(\varepsilon\cdot\exp(-\mathrm{const}/\varepsilon)\). The details are carried out carefully and cannot be included in this review. The analytical results are illustrated by numerical simulations.