no. 16 · Tuesday 17 November 2026
A Microlocal Open-Boundary Method for Residual-Based Wave Solvers on Unbounded Domains
Abstract
We couple the established time-dependent phase-space filter (TDPSF) with a slabwise physics-informed neural network (PINN) for open-domain wave propagation. The neural solver approximates the evolution on each time slab, and localized Fourier multipliers attenuate outgoing buffer components according to their group velocity. The resulting Microlocal PINN Filter (TDPSF-PINN) combines neural residual minimization, sequential buffer filtering, and transfer of the computed endpoint to the next slab. We present $3$ numerical tests: an anisotropic focusing nonlinear Schrödinger packet with a PML comparator, acoustic radiation with a CRBC comparator, and acoustic scattering by three rigid disks with a second-order Bayliss–Turkel comparator. At the reported budgets, TDPSF–PINN shows better accuracy and stability in all 3 tests in the long-time simulations.
- fields
- Numerical Analysis