A perfectly matched layer (PML) is an absorbing region around the simulation interior. Its purpose is to attenuate outgoing waves while introducing little reflection at the entrance to the layer.

FDTD CUDA uses a split-field PML. Inside it, the electric field is reconstructed from two directional contributions:

$$ E_z=E_{zx}+E_{zy} $$

The contributions have separate damping along x and y. Conductivity increases polynomially with depth into the layer, and the magnetic profiles account for the half-cell offsets of the Yee grid. These profiles are included in the update coefficients.

Boundary::Pml enables the layer. Its thickness, polynomial order, and target reflection are configured by pml_thickness, pml_order, and pml_reflection. The layer must leave a nonempty interior, and sources must be outside it.

An outer PEC wall still terminates the grid. Finite thickness and discretization leave some reflection, so the target reflection is a construction parameter rather than a guaranteed measurement. With zero thickness, this option behaves like PEC.