vii
PrefaceAs far as the laws of mathematics refer to reality, they are not certain, and as far as they are certain, they do not refer to reality. Albert EinsteinUsing approximations to find solutions of differential equations has a long history, especially in the areas of applied science and engineering, where relatively simple ways to obtain quantitative results are crucial. Since the 18th century, perturbation methods have been actively used in celestial mechanics for describing planetary motion as a three-body interaction governed by gravitation. As was shown by Poincaré [19], the three-body problem is not completely integrable, and its solution was often based on finding a small perturbation to the solution of the integrable two-body problem, when the mass of at least one of the three bodies is small. One famous example is the prediction of an existing and the position of a new planet by Adams and Leverrier based on calculation of perturbations of Uranus, which resulted in the discovery of the planet Neptune by Galle in 1846. A more recent overview of this topic can be found, for example, in [6,7]. In the 20th century the notion of a "theory of oscillations" as a unifying concept, meaning the application of similar equations and methods of their solution to quite different physical problems, came into being. In particular, the phase plane method and Poincaré mapping proved to be very efficient. For early work in this area, see the classic book by Andronov et al. [3]. viii Preface Regular perturbation schemes allow one to find a deviation from the basic, unperturbed solution, when the deviation remains sufficiently small indefinitely or at least at a time interval at which the solution should be determined. However, in many important cases, a small initial perturbation can strongly change the solution even if the basic equations are only slightly disturbed. In these cases the smallness of the expansion parameter is reflected in the slowness, but not necessarily smallness, of the deviation from the solution existing when that parameter is zero. The slowness means that the time of significant deviation from the unperturbed solution is much longer than the characteristic time scale of the process, such as the period of oscillations or the duration of a pulse. An adequate tool for solving such problems is the asymptotic perturbation theory, which constructs a series in a small parameter in which the main term of the expansion differs from the unperturbed solution in that it contains slowly varying parameters; their variation can be found from "compatibility," or "orthogonality," conditions which secure the finiteness of the higher-order perturbations. In the framework of an asymptotic method, the series may not even converge, but a sum of a finite number of terms in the series approaches the exact solution when the expansion parameter tends to zero, which is sufficient in most applications. Still more involved is singular perturbation theory, in which the asymptotic series does not converge to the unperturbed so...