Path-regularization of linear neutral delay differential equations with several delays (Q495113)

This work considers a class of systems of neutral delay-differential equations with state-dependent delays: \[ \dot{y}(t) = c(y(t)) + \sum_{j=1}^m A_j(y(t)) \dot{y}(\alpha_j(y(t)), \; t>0. \] Even with a smooth function \(y(t)\) for \(t \leq 0\), the derivative of \(y(t)\) for \(t=0\) may be discontinuous and classic solutions to this initial value problem may not exist. Therefore, the authors define a concept of weak solution, based on a differential-algebraic equation, which is a variant of solutions of Utkin type; they also provide a corresponding approximation (``path-regularization'') process. This approach introduces extra parameters that control the speed of change of each component of \(\dot{y}\) in the presence of discontinuities. They can be leveraged to damp oscillations in numerical experiments and also provide a unique solution to the initial value problem, which is not ensured in the Utkin formulation for several delay functions \(\alpha_j\). However, these extra degrees of freedom are not merely design parameters that can be tuned to achieve the best numerical behavior: the definition of the weak solution depends on their choice; it may therefore differ from the solution of Utkin type even in the non-ambiguous case. The authors also compare the asymptotic properties of their approximation process with the exact solutions and then characterize the stability of an associated dynamical system in the case of two independent delay functions.