Wave motion simulation using spectral elements and a hybrid PML formulation

We are concerned with forward wave motion simulations in two-dimensional elastic, heterogeneous, semi-infinite media. We use Perfectly Matched Layers (PMLs) to truncate the semi-infinite extent of the physical domain to arrive at a finite computational domain. We use a recently developed hybrid formulation, where the Navier equations for the interior domain are coupled with a mixed formulation for an unsplit-field PML. Here, we implement the hybrid formulation using spectral elements, and report on its performance. The motivation stems from the following considerations: Of concern is the long-time instability that has been reported even in homogeneous and isotropic cases, when the standard complex-stretching function is used in the PML. The onset of the instability is always within the PML zone, and it manifests as error growth in time. It has been suggested that the instability arises when waves impinge at grazing angle on the PML-interior domain interface. Yet, the instability does not always appear. Furthermore, different values of the various PML parameters (mesh density, attenuation strength, order of attenuation function, etc) can either hinder or delay the onset of the instability. It is thus conjectured that the instability is associated with the spectral properties of the discrete operators. In this thesis, we report numerical results based on both Lagrange interpolants, and results based on spectral elements. Spectral elements are explored since they lead to diagonal mass matrices, have improved dispersion error, and, more importantly, have different spectral properties than Lagrangian-based finite elements. Spectral elements are thus used in an attempt to explore whether the reported instability issues could be alleviated. We design numerical experiments involving explosive sources situated at varying depths from the surface, capable of inducing grazing-angle waves. We use the energy decay as the primary metric for reporting the results of comparisons between various spectral element orders and classical Lagrange interpolants. We also report the results of parametric studies. Overall, it is shown that the spectral elements alone are not capable of removing the instability, though, on occasion, they can. Careful parameterization of the PML could also either remove it or alleviate it. The issue remains open.