Presented here is a time-domain finite element method for approximating Maxwell's equations. The problem is to approximate the electromagnetic fields scattered by a bounded, inhomogeneous cavity embedded in an infinite ground plane. The time-dependent scattering problem is first discretized in time by Newmark's time-stepping scheme. The resulting semidiscrete problem is proved to be well posed. A nonlocal boundary condition on the cavity aperture is constructed to reduce the computational domain to the cavity itself. Stability analysis and error estimates of the fully discrete problem are provided.
Introduction.Time-harmonic (frequency-domain) Maxwell's equations for scattering problems are well studied and documented ([1, 19, 20, 21], to name a few) compared to their time-domain counterparts. This is due, to a large extent, to their obvious advantage of the absence of the time dependence and the limitations of computer power. The recent rapid advances in computing technology have prompted a growing popularity of numerical schemes for simulating electromagnetic transients (time-domain) for their potential to generate wide-band data and model nonlinear materials. Reports of new and faster numerical techniques for electromagnetic analysis have flourished in the engineering literature (see, for example, [18,23,12,25]). However, very little analysis is known in the open literature. To the authors' knowledge, the first mathematical study of time-domain Maxwell's equations for scattering by a bounded perfectly electric conducting (PEC) body was reported in [22], in which a spatially discretized problem is analyzed. More recently, in [4,5], fully discrete finite element methods for solving Maxwell's equations of bounded PEC scattering bodies are considered. In both cases, the problem is defined in a bounded domain. This paper serves as our first attempt to understand the various stability and convergence issues associated with the finite element method for modelling transient electromagnetic scattering from non-PEC bodies. The problem is defined in an infinite space.As observed in [7], most numerical schemes for scattering problems are faced with the problem of truncating the infinite domain to a bounded computational domain