The Helmholtz equation describes how waves, such as sound or light, propagate through a medium. When simulating such waves on a computer, you can only work with a finite region of space, but waves naturally travel outward forever. A standard fix is to surround the computational domain with an artificial absorbing layer called a perfectly matched layer (PML), which is designed to soak up outgoing waves without reflecting them back. This paper studies a particular flavor of PML, the Cartesian version, applied to situations where the wave speed varies from place to place inside the region of interest, for example due to layered materials or embedded objects with different properties.
The mathematical challenge is proving that this setup is both well-posed (meaning it has a unique, stable solution) and that the PML solution actually converges to the true physical solution as the absorbing layer gets thicker. Previous analyses mostly assumed the wave speed was uniform or at most piecewise uniform. Here the authors handle a much broader class of variable speeds, including discontinuous ones, with only minimal assumptions. A key insight is that the variable wave speed only affects lower-order terms in the equations, leaving the dominant structure of the PML unchanged. The authors also identify precisely when wave trapping can occur: if the gradient of the wave speed points inward toward a reference point everywhere, no trapping happens. When this condition fails, waves can get stuck circling inside the medium, which is a genuine physical and mathematical complication.
The main results confirm that the PML problem is well-posed for all wave frequencies, with stability that does not deteriorate as the absorbing layer grows thicker. Most importantly, the error between the PML solution and the true solution shrinks exponentially fast as the layer thickness or the wave frequency increases, which is exactly the property that makes PMLs so attractive in practice. Numerical experiments with a low-velocity inclusion embedded in the domain confirm this exponential convergence, validating the theory and suggesting the approach is robust for realistic, heterogeneous wave propagation problems.