arXiv Science⌕ Search

arXiv · 2611.19603

NAP-Wave: A Neural Adaptive Plane-Wave Architecture for High-Frequency Helmholtz Problems

Abstract

Training neural networks to simulate highly oscillatory wavefields presents optimization challenges, often hindered by spectral bias and poorly conditioned physics-based loss landscapes. To address these challenges and enhance convergence, we propose NAP-Wave, a joint neural architecture and training strategy that expands the target scattered wavefield in an adaptive neural plane-wave basis. The basis functions are extracted from all hidden layers and modulated by trainable plane-wave carriers, embedding the high-frequency oscillatory physics directly into the network's trial space. NAP-Wave is trained on a physics-informed loss built from the Lippmann-Schwinger integral equation, so no differentiation of the network with respect to its input is required. We develop a decoupled optimization strategy in which Adam updates the parameters of the neural plane-wave basis, while the linear combination coefficients are computed exactly by a least-squares (LS) solver. We couple this with an adaptive-depth training strategy in which the network starts shallow and grows progressively layer-by-layer. At each growth step, we avoid the loss spikes typical of dynamic network growth using the LS-computed coefficients, the warm-started parameters, and the preservation of the first- and second-order optimizer moments of the previous stage. Ablation studies confirm that the plane-wave basis, the exact coefficient solve, and the progressive schedule independently accelerate convergence and improve accuracy. Validated on exact-solution benchmarks, NAP-Wave achieves relative L2 errors of order 1%. On highly heterogeneous geophysical benchmarks, it achieves relative errors of 5%, 9%, and 10%, significantly outperforming published PINN methods in both accuracy and convergence time. To ensure reproducibility, the code is available at https://github.com/Hamza-Kamil/NAP-WAVE.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Hamza Kamil, Mohammad Mahdi Abedi, David Pardo. 2026-11-14. NAP-Wave: A Neural Adaptive Plane-Wave Architecture for High-Frequency Helmholtz Problems. https://arxiv.science/abs/2611.19603

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

PyOTIS: an open-source Python package for stream solute transport with transient storage

OTIS and OTIS-P provide transient-storage transport and parameter estimation, whereas E-OTIS adds coupled water-quality reactions in a separate MATLAB implementation. PyOTIS brings these workflows into an open-source Python package with legacy OTIS input support. Its contribution is software integration: a scriptable interface, structured simulation outputs, Jacobian-based parameter uncertainty, sensitivity analysis, and selectable transport schemes. A browser-based graphical user interface (GUI) provides example loading, custom parameter entry, breakthrough-curve and storage-zone plots, reference overlays, and calibration of supplied cases with confidence intervals. The package retains the OTIS Crank--Nicolson formulation and the E-OTIS reaction network; these are inherited model components rather than new process formulations. Evaluation distinguishes code comparisons, analytical convergence, and fits to observations. For the six OTIS manual examples, the comparisons distinguish model outputs from field observations and include agreement to seven significant figures for the steady-state iron profile. The nine-species E-OTIS comparison reports maximum absolute differences of order $10^{-14}$\,mg\,L$^{-1}$ against MATLAB. The conservative analytical benchmark shows approximately second-order convergence. For Clackamas River calibration, residual sums of squares differ from OTIS-P by 3.65--5.12\%, although individual parameter estimates differ by up to 39.1\%. These results support implementation consistency for the tested cases without establishing unique parameter estimates or validating every reaction pathway. Source code, example inputs, and verification scripts accompany the package.

physics.geo-ph↗

2.5D Transformer: An Efficient 3D Seismic Interpolation Method without Full 3D Training

Transformer has emerged as a powerful deep-learning technique for two-dimensional (2D) seismic data interpolation, owing to its global modeling ability. However, its core operation introduces heavy computational burden due to the quadratic complexity, hindering its further application to higher-dimensional data. To achieve Transformer-based three-dimensional (3D) seismic interpolation, we propose a 2.5-dimensional Transformer network (T-2.5D) that adopts a cross-dimensional transfer learning (TL) strategy, so as to adapt the 2D Transformer encoders to 3D seismic data. The proposed T-2.5D is mainly composed of 2D Transformer encoders and 3D seismic dimension adapters (SDAs). Each 3D SDA is placed before a Transformer encoder to learn spatial correlation information across seismic lines. The proposed cross-dimensional TL strategy comprises two stages: 2D pre-training and 3D fine-tuning. In the first stage, we optimize the 2D Transformer encoders using a large amount of 2D data patches. In the second stage, we freeze the 2D Transformer encoders and fine-tune the 3D SDAs using limited 3D data volumes. Extensive experiments on multiple datasets are conducted to assess the effectiveness and efficiency of T-2.5D. Experimental results demonstrate that the proposed method achieves comparable performance to that of full 3D Transformer at a significantly low cost.

physics.geo-ph↗

RevealLO: Global-Scale FWI Including Long-Period Major-Arc Surface Wave and Multi-Orbit Body Wave Data

Global full-waveform inversion (FWI) has matured significantly in recent years, but resolution in the southern hemisphere often remains limited due to sparse receiver coverage. To address this problem, we constructed RevealLO, a global-scale FWI model that updates its predecessor REVEAL by expanding the inversion time window from 1 hour to 3.5 hours. The extended time window incorporates both minor-arc ($R_1, G_1$) and major-arc ($R_2, G_2$) surface wave arrivals, as well as late-arriving multi-orbit body waves, thereby increasing azimuthal coverage beneath ocean basins and the southern hemisphere. We simulate 3-D viscoelastic wavefields using the spectral-element solver Salvus on global fourth-order cubed-sphere meshes, and update velocity structure by successively broadening the period band from 100 - 200 s to 60 - 200 s. Measurements and adjoint sources are based on time-frequency phase misfits, which are suitable for both large-amplitude surface and low-amplitude multiply-reflected body waves. For iterative optimization, we employ a mini-batch stochastic trust-region L-BFGS algorithm, followed by mesh reprojection onto the $33~s$ REVEAL baseline mesh and lowermost-mantle regularization. RevealLO preserves short-period whole-Earth body wave fits while improving azimuthal sampling and mitigating trade-offs between unmodeled azimuthal anisotropy and isotropic structure in the upper and mid mantle.

physics.geo-ph↗