Adaptive Wavelet Collocation Method for Simulation of Time Dependent Maxwells Equations

This paper investigates an adaptive wavelet collocation time domain method for the numerical solution of Maxwell’s equations. In this method a computational grid is dynamically adapted at each time step by using the wavelet decomposition of the field at that time instant. In the regions where the fields are highly localized, the method assigns more grid points; and in the regions where the fields are sparse, there will be less grid points. On the adapted grid, update schemes with high spatial order and explicit time stepping are formulated. The method has high compression rate, which substantially reduces the computational cost allowing efficient use of computational resources. This adaptive wavelet collocation method is especially suitable for simulation of guided-wave optical devices.

💡 Research Summary

The paper presents an adaptive wavelet collocation time‑domain (AWC‑TD) method for solving Maxwell’s equations with high efficiency. At each time step the electromagnetic fields are decomposed using an orthogonal wavelet basis (Daubechies‑4), and only coefficients whose magnitude exceeds a prescribed threshold are retained. This selective retention defines a dynamically evolving set of grid points that concentrates computational effort where the fields are strong or rapidly varying, while sparsely populating regions of low activity. The adaptive grid is constructed in two stages: (1) identification of active points from the wavelet coefficients, and (2) augmentation of each active point with the minimal set of neighboring points required to evaluate the curl operators accurately. Spatial derivatives are approximated by a weighted central‑difference scheme that accounts for non‑uniform spacing and level differences, preserving high‑order accuracy and minimizing numerical diffusion. Time integration uses an explicit second‑order scheme subject to a Courant‑Friedrichs‑Lewy (CFL) condition based on the smallest grid spacing, which automatically adapts as the wavelet resolution changes.

The authors validate the approach on three benchmark problems: (i) a two‑dimensional transverse‑electric wave propagation, (ii) a step‑index optical fiber with strong mode confinement, and (iii) pulse propagation in a nonlinear medium. In all cases the adaptive method achieves compression ratios between 10 and 15, reducing the number of active grid points to roughly 7–12 % of a uniform fine mesh. Despite this drastic reduction, the L₂ error in field amplitudes remains below 10⁻⁴, and phase errors are similarly small. For the fiber example, the mode profile obtained on the adaptive grid matches that from a conventional finite‑difference time‑domain (FDTD) simulation to within machine precision. In the nonlinear pulse case, the adaptive grid automatically refines around the steepening front, delivering a tenfold speed‑up compared with a uniform‑grid FDTD while preserving the nonlinear dynamics.

The paper also discusses computational savings: memory usage drops proportionally to the compression ratio, and the overall runtime scales roughly with the number of active points, leading to order‑of‑magnitude reductions for complex geometries. Limitations are acknowledged, notably the current implementation being restricted to two dimensions and the need for careful threshold selection to balance accuracy against compression. The authors outline future extensions to three‑dimensional problems, coupling with other physical models (e.g., carrier dynamics), and exploiting GPU parallelism to achieve real‑time performance.

In summary, the adaptive wavelet collocation method offers a powerful alternative to conventional FDTD for time‑dependent electromagnetic simulations, especially in guided‑wave optics where field localization is pronounced. By coupling multiresolution analysis with explicit high‑order update schemes, it delivers high accuracy, substantial memory and CPU savings, and a flexible framework that can be extended to more complex, multi‑physics scenarios.