What is the PML absorbing boundary condition and how does it work in FDTD simulation?
PML Theory and Configuration
Before PML's invention by Jean-Pierre Berenger in 1994, FDTD simulations of open-boundary problems suffered from significant artificial reflections from the domain boundary. PML revolutionized computational electromagnetics by enabling accurate simulation of antennas, scattering, and radiation problems in finite computational domains.
PML Theory
PML introduces an anisotropic absorbing medium where the conductivity is applied independently in each coordinate direction. A wave propagating in the x-direction encounters x-directed conductivity sigma_x, which attenuates it exponentially: E = E_0 × exp(-sigma_x × x / (2 × epsilon_0 × c)). The mathematical trick is that the wave impedance remains equal to the adjacent medium (eta = sqrt(mu/epsilon) is unchanged because both mu and epsilon are modified by the same complex stretch factor), so there is zero reflection at the interface in the continuous formulation. The complex coordinate stretching formulation: x → x_tilde = x × (1 + sigma_x / (j × omega × epsilon_0)), transforms the real coordinate into a complex coordinate that causes exponential decay. This formulation is more general and numerically stable than the original split-field approach.
Configuration Guidelines
For FDTD simulation: (1) PML thickness: 8 cells for quick simulations, 12-16 cells for high accuracy. Each additional cell reduces reflection by approximately 6-10 dB. (2) Grading order: cubic polynomial (m=3) for the conductivity profile: sigma(d) = sigma_max × (d/d_total)^3, where d is distance from the inner PML boundary. (3) Maximum conductivity: sigma_max = -(m+1) × ln(R_0) / (2 × eta × d_total), where R_0 is the desired reflection coefficient at normal incidence (typically 10^-6 to 10^-8) and eta is the wave impedance. (4) Distance to objects: maintain at least lambda/4 (and preferably lambda/2) between any radiating or scattering object and the PML inner boundary. Objects too close to PML couple evanescent near-fields into the absorber, causing numerical artifacts. (5) Cornersand edges: PML regions overlap at simulation domain corners and edges, where two or three PML directions are simultaneously active. Modern implementations handle this automatically, but older codes may require explicit corner treatment.
Common PML Problems
(1) Late-time instability: some PML formulations (particularly the original Berenger split-field PML) can develop exponentially growing fields after many thousands of time steps. The CPML (Convolutional PML) formulation eliminates this instability and is the recommended implementation for all modern FDTD codes. (2) Low-frequency breakdown: standard PML becomes less effective at very low frequencies (electrically small domains) because the PML conductivity profile is optimized for propagating waves. The CFS-PML (Complex Frequency-Shifted PML) adds a real-valued shift parameter that improves absorption at low frequencies and for evanescent waves. (3) Anisotropic media adjacent to PML: when the simulation contains anisotropic materials (ferrites, magnetized plasmas) touching the PML boundary, the PML impedance match may not hold, causing reflections. Solution: add a buffer of isotropic material between the anisotropic medium and PML. (4) Waveguide modes in PML: PML can absorb guided modes that should propagate through a waveguide that exits the simulation domain. Use waveguide ports instead of PML at waveguide terminations.
σ_max = -(m+1)·ln(R₀) / (2·η·d_total)
R(θ) = exp(-2·σ_max·d·cos(θ)/(ε₀·c·(m+1)))
Minimum PML-object distance: λ/4 to λ/2
Frequently Asked Questions
How many PML cells do I need?
8 cells minimum for engineering accuracy (-40 to -50 dB reflection). 12 cells for high accuracy (-60 to -70 dB). 16 cells for precision (-70 to -80 dB). Each doubling of PML cells reduces reflection by approximately 20 dB but doubles the PML memory and increases total simulation domain size. For most antenna and scattering problems, 10-12 cells with cubic grading provides adequate accuracy. For near-field calculations very close to PML (e.g., evanescent wave analysis), use 16+ cells and CFS-PML formulation.
What is the difference between PML and ABC?
Absorbing Boundary Conditions (ABCs) like Mur or Liao are single-layer boundary conditions applied at the domain edge that approximate the radiation condition. They are simpler (no additional cells needed) but provide limited absorption (-20 to -30 dB) and degrade significantly at non-normal incidence angles. PML is a multi-cell absorbing region that provides much better absorption (-40 to -80 dB) at all angles. PML has completely replaced ABCs in modern FDTD codes. The only remaining use for ABCs is in educational codes or extremely memory-constrained situations where the 8-16 cell PML overhead is unacceptable.
Can PML absorb DC or static fields?
Standard PML does not absorb DC (zero frequency) fields because the conductivity-based absorption requires propagating wave behavior. The CFS-PML formulation includes a real-valued frequency-shift parameter alpha that provides absorption even at DC. This is important for simulations involving charged particles, static field problems, or very low-frequency transient responses. Without CFS-PML, DC or very-low-frequency components reflect from the PML boundary and can cause non-physical late-time fields that corrupt the simulation results.