Searcharxiv⌕ Search

arXiv subjects

Zachary E. Ross

Publications and source records attributed to Zachary E. Ross.

At least 19 recordsLinked to original sources
arXiv2026

Forward and Inverse Mantle Convection with Neural Operators

Thermal state reconstruction--reversing convection to recover the thermal structure of the mantle at an earlier geologic time--is an important tool to understand the evolution of mantle convection and its relation to seismic tomographic images and observations at the surface. Thermal state reconstructions are computationally expensive. Here we transformed the basic computational element, numerical solvers, into neural operators, a class of machine learning models for learning mappings between function spaces. Focusing on a specific architecture, Fourier Neural Operators, we demonstrate that they can represent not only a surrogate model like the Stokes system of equations using a purely physics informed approach, but also discover operators without explicit mathematical formulations or even ill-posedness from data, including the direct mapping between two convecting thermal states separated by a long time interval much larger than the Courant-Friedrichs-Lewy condition and its reversal. These neural operators significantly accelerate forward and inverse convection modelling by transforming forward physical processes into surrogate models with lower complexity while utilizing auto-differentiation to calculate gradients. With this framework, we demonstrate the strengths and weaknesses of four methods for thermal state reconstructions: reverse buoyancy, reverse convection operator, an inversion with only the terminal thermal state, and a joint inversion with the terminal thermal state and surface velocity evolution. The reverse convection operator is shown to perform poorly in the presence of observational noise, but the joint inversion overcomes this limitation. The joint technique could probably become a solution to large-scale thermal state inversion problems using seismic tomography and plate tectonic reconstructions.

physics.geo-ph↗
arXiv2026

Flow Annealing Posterior Sampling for Function-Space Regression and Inverse Problems

Principled regression for stochastic processes is a long-standing challenge with deep connections to scientific inverse problems. We introduce Flow Annealing Posterior Sampling (FAPS), to our knowledge the first function-space posterior sampling framework that unifies stochastic-process regression and PDE inverse problems. Built on pretrained function-space flow-matching priors, FAPS enables likelihood-guided posterior inference from sparse and noisy observations, supports variable query discretizations, and avoids explicit prior-density evaluation. Its Langevin correction uses a low-rank covariance preconditioner to exploit dominant function-space correlations across discretizations. Across Gaussian and non-Gaussian stochastic-process regression benchmarks and diverse PDE inverse problems, FAPS produces coherent posterior samples with accurate uncertainty quantification, significantly outperforming existing functional regression baselines and achieving competitive or better PDE noisy inverse performance than diffusion-based posterior samplers while reducing test-time sampling cost.

stat.ML↗
arXiv2026

Enforcing Reciprocity in Operator Learning for Seismic Wave Propagation

Accurate and efficient wavefield modeling underpins seismic structure and source studies. Traditional methods comply with physical laws but are computationally intensive. Data-driven methods, while opening new avenues for advancement, have yet to incorporate strict physical consistency. The principle of reciprocity is one of the most fundamental physical laws in wave propagation. We introduce the Reciprocity-Enforced Neural Operator (RENO), a transformer-based architecture for modeling seismic wave propagation that hard-codes the reciprocity principle. The model leverages the cross-attention mechanism and commutative operations to guarantee invariance under swapping source and receiver positions. Beyond improved physical consistency, the proposed architecture supports simultaneous realizations for multiple sources. This yields an order-of-magnitude inference speedup at a similar memory footprint over a conventional neural operator on a realistic multi-source configuration. We demonstrate the functionality using the reciprocity relation for particle velocity fields under single forces. This architecture is also applicable to pressure fields under dilatational sources and travel-time fields governed by the eikonal equation, paving the way for encoding more complex reciprocity relations.

physics.geo-ph↗
arXiv2026

Large-Scale 3D Ground-Motion Synthesis with Physics-Inspired Latent Operator Flow Matching

Earthquake hazard analysis and design of spatially distributed infrastructure, such as power grids and energy pipeline networks, require scenario-specific ground-motion time histories with realistic frequency content and spatiotemporal coherence. However, producing the large ensembles needed for uncertainty quantification with physics-based simulations is computationally intensive and impractical for engineering workflows. To address this challenge, we introduce Ground-Motion Flow (GMFlow), a physics-inspired latent operator flow matching framework that generates realistic, large-scale regional ground-motion time-histories conditioned on physical parameters. Validated on simulated earthquake scenarios in the San Francisco Bay Area, GMFlow generates spatially coherent ground motion across more than 9 million grid points in seconds, achieving a 10,000-fold speedup over the simulation workflow, which opens a path toward rapid and uncertainty-aware hazard assessment for distributed infrastructure. More broadly, GMFlow advances mesh-agnostic functional generative modeling and could potentially be extended to the synthesis of large-scale spatiotemporal physical fields in diverse scientific domains.

cs.LG↗
arXiv2026

SPIDER: Scalable Probabilistic Inference for Differential Earthquake Relocation

Seismicity catalogs are larger than ever due to an explosion of techniques for enhanced earthquake detection and an abundance of high-quality datasets. Bayesian inference is an appealing framework for locating earthquakes due to its ability to propagate and quantify uncertainty into the inversion results, but traditional methods do not scale well to high-dimensional parameter spaces, making them unsuitable for double-difference relocation where the number of parameters can reach the millions. Here we introduce SPIDER, a scalable Bayesian inference framework for double-difference hypocenter relocation. SPIDER uses a physics-informed neural network Eikonal solver together with a highly efficient sampler called Stochastic Gradient Langevin Dynamics to generate posterior samples jointly for entire seismicity catalogs. We show that traditional double-difference relocation formulations neglect residual correlation between observations with common events, which biases uncertainty estimates. Our formulation is designed to whiten this residual correlation, and is readily parallelized over multiple GPUs for enhanced computational efficiency. We demonstrate the capabilities of SPIDER on a rigorous synthetic seismicity catalog and three real data catalogs from California and Japan. We introduce several ways to analyze high-dimensional posterior distributions to aid in scientific interpretation and evaluation.

physics.geo-ph↗
arXiv2025

Ambient Noise Full Waveform Inversion with Neural Operators

Numerical simulations of seismic wave propagation are crucial for investigating velocity structures and improving seismic hazard assessment. However, standard methods such as finite difference or finite element are computationally expensive. Recent studies have shown that a new class of machine learning models, called neural operators, can solve the elastodynamic wave equation orders of magnitude faster than conventional methods. Full waveform inversion is a prime beneficiary of the accelerated simulations. Neural operators, as end-to-end differentiable operators, combined with automatic differentiation, provide an alternative approach to the adjoint-state method. State-of-the-art optimization techniques built into PyTorch provide neural operators with greater flexibility to improve the optimization dynamics of full waveform inversion, thereby mitigating cycle-skipping problems. In this study, we demonstrate the first application of neural operators for full waveform inversion on a real seismic dataset, which consists of several nodal transects collected across the San Gabriel, Chino, and San Bernardino basins in the Los Angeles metropolitan area.

physics.geo-ph↗
arXiv2025

Mesh-Informed Neural Operator : A Transformer Generative Approach

Generative models in function spaces, situated at the intersection of generative modeling and operator learning, are attracting increasing attention due to their immense potential in diverse scientific and engineering applications. While functional generative models are theoretically domain- and discretization-agnostic, current implementations heavily rely on the Fourier Neural Operator (FNO), limiting their applicability to regular grids and rectangular domains. To overcome these critical limitations, we introduce the Mesh-Informed Neural Operator (MINO). By leveraging graph neural operators and cross-attention mechanisms, MINO offers a principled, domain- and discretization-agnostic backbone for generative modeling in function spaces. This advancement significantly expands the scope of such models to more diverse applications in generative, inverse, and regression tasks. Furthermore, MINO provides a unified perspective on integrating neural operators with general advanced deep learning architectures. Finally, we introduce a suite of standardized evaluation metrics that enable objective comparison of functional generative models, addressing another critical gap in the field.

cs.LG↗
arXiv2025

Stochastic Process Learning via Operator Flow Matching

Expanding on neural operators, we propose a novel framework for stochastic process learning across arbitrary domains. In particular, we develop operator flow matching (OFM) for learning stochastic process priors on function spaces. OFM provides the probability density of the values of any collection of points and enables mathematically tractable functional regression at new points with mean and density estimation. Our method outperforms state-of-the-art models in stochastic process learning, functional regression, and prior learning.

cs.LG↗
arXiv2025

InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences

Plug-and-play diffusion priors (PnPDP) have emerged as a promising research direction for solving inverse problems. However, current studies primarily focus on natural image restoration, leaving the performance of these algorithms in scientific inverse problems largely unexplored. To address this gap, we introduce \textsc{InverseBench}, a framework that evaluates diffusion models across five distinct scientific inverse problems. These problems present unique structural challenges that differ from existing benchmarks, arising from critical scientific applications such as optical tomography, medical imaging, black hole imaging, seismology, and fluid dynamics. With \textsc{InverseBench}, we benchmark 14 inverse problem algorithms that use plug-and-play diffusion priors against strong, domain-specific baselines, offering valuable new insights into the strengths and weaknesses of existing algorithms. To facilitate further research and development, we open-source the codebase, along with datasets and pre-trained models, at https://devzhk.github.io/InverseBench/.

cs.LG↗
arXiv2025

High-resolution eikonal imaging and uncertainty quantification of the Kilauea caldera

Images of the Earth's interior can provide us with insight into the underlying properties of the Earth, such as how seismic activity might emerge and the interplay between seismic and volcanic activity. Understanding these systems requires reliable high-resolution images to understand mechanisms and estimate physical quantities. However, reliable images are often difficult to obtain due to the non-linear nature of seismic wave propagation and the ill-posedness of the related inverse problem. Reconstructions rely on good initial estimates as well as hand-crafted priors, which can ultimately bias solutions. In our work, we present a 3D reconstruction of Kilauea's magmatic system at a previously unattained resolution. Our eikonal tomography procedure improves upon prior imaging results of Kilauea through increased resolution and per-pixel uncertainties estimated through variational inference. In particular, solving eikonal imaging using variational inference with stochastic gradient descent enables stable inversion and uncertainty quantification in the absence of strong prior knowledge of the velocity structure. Our work makes two key contributions: developing a stochastic eikonal tomography scheme with uncertainty quantification and illuminating the structure and melt quantity of the magmatic system that underlies Kilauea.

physics.geo-ph↗
arXiv2025

Reducing Frequency Bias of Fourier Neural Operators in 3D Seismic Wavefield Simulations Through Multi-Stage Training

The recent development of Neural Operator (NeurOp) learning for solutions to the elastic wave equation shows promising results and provides the basis for fast large-scale simulations for different seismological applications. In this paper, we use the Fourier Neural Operator (FNO) model to directly solve the 3D Helmholtz wave equation for fast seismic ground motion simulations on different frequencies, and show the frequency bias of the FNO model, i.e. it learns the lower frequencies better comparing to the higher frequencies. To reduce the frequency bias, we adopt the multi-stage FNO training, i.e., after training a 1st stage FNO model for estimating the ground motion, we use a second FNO model as the 2nd stage to learn from the residual, which greatly reduced the errors on the higher frequencies. By adopting this multi-stage training, the FNO models show reduced biases on higher frequencies, which enhanced the overall results of the ground motion simulations. Thus the multi-stage training FNO improves the accuracy and realism of the ground motion simulations.

physics.geo-ph↗
arXiv2024

Universal Functional Regression with Neural Operator Flows

Regression on function spaces is typically limited to models with Gaussian process priors. We introduce the notion of universal functional regression, in which we aim to learn a prior distribution over non-Gaussian function spaces that remains mathematically tractable for functional regression. To do this, we develop Neural Operator Flows (OpFlow), an infinite-dimensional extension of normalizing flows. OpFlow is an invertible operator that maps the (potentially unknown) data function space into a Gaussian process, allowing for exact likelihood estimation of functional point evaluations. OpFlow enables robust and accurate uncertainty quantification via drawing posterior samples of the Gaussian process and subsequently mapping them into the data function space. We empirically study the performance of OpFlow on regression and generation tasks with data generated from Gaussian processes with known posterior forms and non-Gaussian processes, as well as real-world earthquake seismograms with an unknown closed-form distribution.

cs.LG↗
arXiv2024

Insights on the dip of fault zones in Southern California from modeling of seismicity with anisotropic point processes

Accurate models of fault zone geometry are important for scientific and hazard applications. While seismicity can provide high-resolution point measurements of fault geometry, extrapolating these measurements to volumes may involve making strong assumptions. This is particularly problematic in distributed fault zones, which are commonly observed in immature faulting regions. In this study, we focus on characterizing the dip of fault zones in Southern California with the goal of improving fault models. We introduce a novel technique from spatial point process theory to quantify the orientation of persistent surficial features in seismicity, even when embedded in wide shear zones. The technique makes relatively mild assumptions about fault geometry and is formulated with the goal of determining the dip of a fault zone at depth. The method is applied to 11 prominent seismicity regions in Southern California. Overall, the results compare favorably with the geometry models provided by the SCEC Community Fault Model and other focused regional studies. More specifically, we find evidence that the Southern San Andreas and San Jacinto fault zones are both northeast dipping at seismogenic depths at the length scales of 1.0-4.0 km. In addition, we find more limited evidence for some depth dependent variations in dip that suggest a listric geometry. The developed technique can provide an independent source of information from seismicity to augment existing fault geometry models.

physics.geo-ph↗
arXiv2024

Broadband Ground Motion Synthesis via Generative Adversarial Neural Operators: Development and Validation

We present a data-driven framework for ground-motion synthesis that generates three-component acceleration time histories conditioned on moment magnitude, rupture distance , time-average shear-wave velocity at the top $30m$ ($V_{S30}$), and style of faulting. We use a Generative Adversarial Neural Operator (GANO), a resolution invariant architecture that guarantees model training independent of the data sampling frequency. We first present the conditional ground-motion synthesis algorithm (cGM-GANO) and discuss its advantages compared to previous work. We next train cGM-GANO on simulated ground motions generated by the Southern California Earthquake Center Broadband Platform (BBP) and on recorded KiK-net data and show that the model can learn the overall magnitude, distance, and $V_{S30}$ scaling of effective amplitude spectra (EAS) ordinates and pseudo-spectral accelerations (PSA). Results specifically show that cGM-GANO produces consistent median scaling with the training data for the corresponding tectonic environments over a wide range of frequencies for scenarios with sufficient data coverage. For the BBP dataset, cGM-GANO cannot learn the ground motion scaling of the stochastic frequency components; for the KiK-net dataset, the largest misfit is observed at short distances and for soft soil conditions due to the scarcity of such data. Except for these conditions, the aleatory variability of EAS and PSA are captured reasonably well. Lastly, cGM-GANO produces similar median scaling to traditional GMMs for frequencies greater than 1Hz for both PSA and EAS but underestimates the aleatory variability of EAS. Discrepancies in the comparisons between the synthetic ground motions and GMMs are attributed to inconsistencies between the training dataset and the datasets used in GMM development. Our pilot study demonstrates GANO's potential for efficient synthesis of broad-band ground motions

physics.geo-ph↗
arXiv2023

Phase Neural Operator for Multi-Station Picking of Seismic Arrivals

Seismic wave arrival time measurements form the basis for numerous downstream applications. State-of-the-art approaches for phase picking use deep neural networks to annotate seismograms at each station independently, yet human experts annotate seismic data by examining the whole network jointly. Here, we introduce a general-purpose network-wide phase picking algorithm based on a recently developed machine learning paradigm called Neural Operator. Our model, called PhaseNO, leverages the spatio-temporal contextual information to pick phases simultaneously for any seismic network geometry. This results in superior performance over leading baseline algorithms by detecting many more earthquakes, picking more phase arrivals, while also greatly improving measurement accuracy. Following similar trends being seen across the domains of artificial intelligence, our approach provides but a glimpse of the potential gains from fully-utilizing the massive seismic datasets being collected worldwide.

physics.geo-ph↗
arXiv2023

Deep Neural Helmholtz Operators for 3D Elastic Wave Propagation and Inversion

Numerical simulations of seismic wave propagation in heterogeneous 3D media are central to investigating subsurface structures and understanding earthquake processes, yet are computationally expensive for large problems. This is particularly problematic for full waveform inversion, which typically involves numerous runs of the forward process. In machine learning there has been considerable recent work in the area of operator learning, with a new class of models called neural operators allowing for data-driven solutions to partial differential equations. Recent works in seismology have shown that when neural operators are adequately trained, they can significantly shorten the compute time for wave propagation. However, the memory required for the 3D time domain equations may be prohibitive. In this study, we show that these limitations can be overcome by solving the wave equations in the frequency domain, also known as the Helmholtz equations, since the solutions for a set of frequencies can be determined in parallel. The 3D Helmholtz neural operator is 40 times more memory-efficient than an equivalent time-domain version. We employ a U-shaped neural operator for 2D and 3D elastic wave modeling, achieving two orders of magnitude acceleration compared to a baseline spectral element method. The neural operator accurately generalizes to variable velocity structures and can be evaluated on denser input meshes than used in the training simulations. We also show that when solving for wavefields strictly on the surface, the accuracy can be significantly improved via a graph neural operator layer. In leveraging automatic differentiation, the proposed method can serve as an alternative to the adjoint-state approach for 3D full-waveform inversion, reducing the computation time by a factor of 350.

physics.geo-ph↗
arXiv2023

U-NO: U-shaped Neural Operators

Neural operators generalize classical neural networks to maps between infinite-dimensional spaces, e.g., function spaces. Prior works on neural operators proposed a series of novel methods to learn such maps and demonstrated unprecedented success in learning solution operators of partial differential equations. Due to their close proximity to fully connected architectures, these models mainly suffer from high memory usage and are generally limited to shallow deep learning models. In this paper, we propose U-shaped Neural Operator (U-NO), a U-shaped memory enhanced architecture that allows for deeper neural operators. U-NOs exploit the problem structures in function predictions and demonstrate fast training, data efficiency, and robustness with respect to hyperparameters choices. We study the performance of U-NO on PDE benchmarks, namely, Darcy's flow law and the Navier-Stokes equations. We show that U-NO results in an average of 26% and 44% prediction improvement on Darcy's flow and turbulent Navier-Stokes equations, respectively, over the state of the art. On Navier-Stokes 3D spatiotemporal operator learning task, we show U-NO provides 37% improvement over the state of art methods.

cs.LG↗
arXiv2023

Rapid Seismic Waveform Modeling and Inversion with Neural Operators

Seismic waveform modeling is a powerful tool for determining earth structure models and unraveling earthquake rupture processes, but it is usually computationally expensive. We introduce a scheme to vastly accelerate these calculations with a recently developed machine learning paradigm called the neural operator. Once trained, these models can simulate a full wavefield at negligible cost. We use a U-shaped neural operator to learn a general solution operator to the 2D elastic wave equation from an ensemble of numerical simulations performed with random velocity models and source locations. We show that full waveform modeling with neural operators is nearly two orders of magnitude faster than conventional numerical methods, and more importantly, the trained model enables accurate simulation for velocity models, source locations, and mesh discretization distinctly different from the training dataset. The method also enables convenient full-waveform inversion with automatic differentiation.

physics.geo-ph↗