pith. sign in

arxiv: 2010.08895 · v3 · submitted 2020-10-18 · 💻 cs.LG · cs.NA· math.NA

Fourier Neural Operator for Parametric Partial Differential Equations

Zongyi Li , Nikola Kovachki , Kamyar Azizzadenesheli , Burigede Liu , Kaushik Bhattacharya , Andrew Stuart , Anima Anandkumar This is my paper

Pith reviewed 2026-05-11 00:51 UTC · model grok-4.3

classification 💻 cs.LG cs.NAmath.NA
keywords Fourier neural operatorneural operatorsparametric PDEsBurgers equationDarcy flowNavier-Stokesturbulent flowszero-shot super-resolution
0
0 comments X

The pith

Parameterizing the integral kernel directly in Fourier space creates a neural operator that learns solution mappings for entire families of PDEs with high accuracy and speed.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a neural operator that learns mappings between function spaces rather than finite-dimensional vectors, allowing it to capture an entire family of parametric PDEs at once instead of solving one instance at a time. By directly parameterizing the integral kernel in Fourier space, the architecture computes convolutions efficiently via the FFT while retaining expressivity for complex nonlinear dynamics. Experiments on Burgers' equation, Darcy flow, and the Navier-Stokes equations show the method achieves superior accuracy at fixed resolution compared with prior learning-based solvers and, uniquely among ML approaches, models turbulent flows with zero-shot super-resolution. If the central claim holds, repeated expensive PDE solves across parameter ranges become unnecessary, replacing them with a single trained operator that generalizes across resolutions and parameters.

Core claim

The Fourier neural operator approximates the solution operator of parametric PDEs by representing the integral kernel directly as a learnable function in Fourier space. This yields an architecture that is both expressive enough to capture turbulent regimes and efficient enough to deliver up to three orders of magnitude speedup over classical solvers while enabling zero-shot super-resolution on Navier-Stokes flows.

What carries the argument

The Fourier layer that parameterizes the kernel integral operator directly in Fourier space and evaluates it via the fast Fourier transform.

If this is right

  • The same trained model solves the PDE for any choice of parameters without retraining or re-discretization.
  • Zero-shot super-resolution becomes possible for turbulent flows, producing accurate fine-scale details from coarse training data.
  • Runtime cost drops by up to three orders of magnitude relative to classical PDE solvers for repeated evaluations across parameter space.
  • Accuracy at fixed resolution exceeds that of earlier neural-network-based PDE solvers on the tested families.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same Fourier parameterization could be applied to learn operators for inverse problems or optimal control of PDEs without changing the core architecture.
  • Because the method works in Fourier space, it may naturally extend to periodic domains or problems where spectral methods already dominate.
  • Training on multiple PDE families simultaneously might produce a single model that switches between equation types by changing only the input parameters.

Load-bearing premise

That directly parameterizing the integral kernel in Fourier space supplies enough expressivity and stability for the target PDE families without introducing artifacts or requiring prohibitive amounts of training data.

What would settle it

A demonstration that the trained Fourier neural operator produces visibly incorrect velocity or pressure fields for a turbulent Navier-Stokes test case at a resolution higher than its training data, or that its wall-clock time on equivalent hardware is not substantially lower than a traditional spectral solver.

read the original abstract

The classical development of neural networks has primarily focused on learning mappings between finite-dimensional Euclidean spaces. Recently, this has been generalized to neural operators that learn mappings between function spaces. For partial differential equations (PDEs), neural operators directly learn the mapping from any functional parametric dependence to the solution. Thus, they learn an entire family of PDEs, in contrast to classical methods which solve one instance of the equation. In this work, we formulate a new neural operator by parameterizing the integral kernel directly in Fourier space, allowing for an expressive and efficient architecture. We perform experiments on Burgers' equation, Darcy flow, and Navier-Stokes equation. The Fourier neural operator is the first ML-based method to successfully model turbulent flows with zero-shot super-resolution. It is up to three orders of magnitude faster compared to traditional PDE solvers. Additionally, it achieves superior accuracy compared to previous learning-based solvers under fixed resolution.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces the Fourier Neural Operator (FNO), a neural architecture that learns mappings between function spaces for parametric PDEs by directly parameterizing the integral kernel in Fourier space. Experiments are reported on Burgers' equation, Darcy flow, and the Navier-Stokes equations, with the central claims being that FNO is the first ML-based method to model turbulent flows with zero-shot super-resolution, achieves up to three orders of magnitude speedup over traditional PDE solvers, and attains superior accuracy relative to prior learning-based solvers at fixed resolution.

Significance. If the empirical results hold under more detailed scrutiny, this work would represent a meaningful advance in scientific machine learning by providing a resolution-independent operator learning framework that leverages FFT for efficiency. The ability to handle parametric families and perform zero-shot super-resolution on turbulent flows could accelerate surrogate modeling in fluid dynamics and related domains, building directly on prior neural operator ideas with a computationally favorable parameterization.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (Experiments): The central performance claims for Navier-Stokes (zero-shot super-resolution and three-order-of-magnitude speedup) are only moderately supported because the abstract and experimental section lack explicit details on training protocols, exact baseline implementations, error metrics (e.g., relative L2 norms with confidence intervals), and statistical significance testing across multiple random seeds. This information is load-bearing for validating the superiority and generalization assertions.
  2. [§3.2] §3.2 (Fourier layer): The direct parameterization of the kernel via a finite number of Fourier modes is presented as yielding sufficient expressivity and stability for the energy cascade in turbulent flows, yet no analysis of truncation error, dealiasing strategy, or high-wavenumber fidelity is provided. This is critical because the zero-shot super-resolution claim implicitly assumes that retained modes generalize without spectral bias or instability when resolution increases.
minor comments (1)
  1. [§3] Notation for the Fourier multiplier R in the layer definition could be clarified with an explicit statement of its dependence on the input parameters to avoid ambiguity when comparing to non-parametric baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential of the Fourier Neural Operator framework. We address each major comment below, providing clarifications from the manuscript and indicating revisions where they will strengthen the presentation without altering the core claims.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (Experiments): The central performance claims for Navier-Stokes (zero-shot super-resolution and three-order-of-magnitude speedup) are only moderately supported because the abstract and experimental section lack explicit details on training protocols, exact baseline implementations, error metrics (e.g., relative L2 norms with confidence intervals), and statistical significance testing across multiple random seeds. This information is load-bearing for validating the superiority and generalization assertions.

    Authors: We agree that greater explicitness in the experimental section will improve reproducibility and support for the claims. The abstract is a high-level summary and does not contain implementation details by design. In the revised manuscript we will expand §4 (and the appendix) to include: (i) full training protocols (optimizer, learning-rate schedule, batch size, epochs, and data generation parameters); (ii) precise descriptions of baseline re-implementations with citations to the original codes; (iii) relative L2 error metrics reported with standard deviations computed over multiple random seeds; and (iv) a short discussion of statistical significance where the differences are large. These additions will be placed in the main experimental section rather than only in supplementary material. revision: yes

  2. Referee: [§3.2] §3.2 (Fourier layer): The direct parameterization of the kernel via a finite number of Fourier modes is presented as yielding sufficient expressivity and stability for the energy cascade in turbulent flows, yet no analysis of truncation error, dealiasing strategy, or high-wavenumber fidelity is provided. This is critical because the zero-shot super-resolution claim implicitly assumes that retained modes generalize without spectral bias or instability when resolution increases.

    Authors: The finite-mode parameterization is chosen for both efficiency and because the dominant dynamics of the target PDEs (including the energy cascade in Navier-Stokes) are captured by a modest number of low-to-mid wavenumbers; this is standard practice in spectral numerical methods. The zero-shot super-resolution result is empirical: the operator is trained at one resolution and evaluated at higher resolutions using the same learned Fourier weights, with accuracy verified directly on the test data. While the manuscript does not contain a dedicated theoretical truncation-error analysis, the implementation relies on standard FFT routines whose aliasing properties are well-understood. In revision we will add a concise paragraph in §3.2 that (a) states the mode cutoff used, (b) notes that dealiasing is handled implicitly by the discrete Fourier transform, and (c) provides additional empirical plots of error versus wavenumber to illustrate high-wavenumber fidelity. A full spectral-bias theory lies outside the scope of the present work, which focuses on the operator-learning architecture and its practical performance. revision: partial

Circularity Check

0 steps flagged

No circularity in Fourier Neural Operator architecture or claims

full rationale

The paper defines the Fourier neural operator as a new architecture that directly parameterizes the integral kernel in Fourier space for learning mappings between function spaces. This is an explicit design choice presented in the abstract and full text, followed by empirical experiments on Burgers' equation, Darcy flow, and Navier-Stokes. No derivation step reduces a claimed result or prediction to a fitted parameter or prior self-citation by construction. The zero-shot super-resolution and speed claims are presented as experimental outcomes, not tautological outputs. The contribution remains self-contained as an architectural proposal with independent validation on standard benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, no explicit free parameters, axioms, or invented entities are detailed beyond the architectural choice of Fourier kernel parameterization.

pith-pipeline@v0.9.0 · 5476 in / 1038 out tokens · 63847 ms · 2026-05-11T00:51:55.821593+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 60 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. KAN: Kolmogorov-Arnold Networks

    cs.LG 2024-04 conditional novelty 8.0

    KANs with learnable univariate spline activations on edges achieve better accuracy than MLPs with fewer parameters, faster scaling, and direct visualization for scientific discovery.

  2. Observation-Aligned Mask Priors for Learning Physical Dynamics from Authentic Occlusions

    cs.CV 2026-05 unverdicted novelty 7.0

    A framework pretrained on authentic binary occlusion masks uses guided sampling and intersection-based partitioning to train diffusion models on incomplete physical observations without zero-query regions.

  3. Amortized Energy-Based Bayesian Inference

    math.NA 2026-05 unverdicted novelty 7.0

    Presents a likelihood-free transport map learned by minimizing an averaged energy-distance objective to amortize Bayesian inference for inverse problems, including PDE-constrained cases with neural operator representations.

  4. Discovering Physical Directions in Weight Space: Composing Neural PDE Experts

    cs.LG 2026-05 unverdicted novelty 7.0

    Fine-tuning neural PDE operators to regime endpoints reveals a physical direction in weight space that CCM uses to compose accurate merged models for new or extrapolated regimes from metadata or short prefixes.

  5. Topology-Preserving Neural Operator Learning via Hodge Decomposition

    cs.LG 2026-05 unverdicted novelty 7.0

    Hodge Spectral Duality provides a topology-preserving neural operator by isolating unlearnable topological components via Hodge orthogonality and operator splitting.

  6. Neural-Schwarz Tiling for Geometry-Universal PDE Solving at Scale

    cs.LG 2026-05 unverdicted novelty 7.0

    Local neural operators on 3x3x3 patches, composed via Schwarz iteration, solve large-scale nonlinear elasticity on arbitrary geometries without domain-specific retraining.

  7. Approximation Theory of Laplacian-Based Neural Operators for Reaction-Diffusion System

    cs.LG 2026-05 unverdicted novelty 7.0

    Laplacian eigenfunction-based neural operators approximate the solution operator of the generalized Gierer-Meinhardt reaction-diffusion system with error bounds that imply only polynomial growth in parameters as accur...

  8. Fixed-Point Neural Optimal Transport without Implicit Differentiation

    math.OC 2026-05 unverdicted novelty 7.0

    A single-network fixed-point formulation for neural optimal transport eliminates adversarial min-max optimization and implicit differentiation while enforcing dual feasibility exactly.

  9. Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators

    cs.LG 2026-05 unverdicted novelty 7.0

    A latent Structured Spectral Propagator enables stable autoregressive PDE forecasting by decoupling spatial details from recurrent modal dynamics.

  10. Your Simulation Runs but Solves the Wrong Physics: PDE-Grounded Intent Verification for LLM-Generated Multiphysics Simulation Code

    cs.LG 2026-05 unverdicted novelty 7.0

    A new Intent Fidelity Score and refinement loop verify that LLM-generated simulation code matches the intended PDEs, improving performance on a 220-case benchmark where execution alone fails to ensure correctness.

  11. CATO: Charted Attention for Neural PDE Operators

    cs.AI 2026-05 unverdicted novelty 7.0

    CATO learns a continuous latent chart for efficient axial attention on PDE meshes and adds derivative-aware supervision to improve accuracy and reduce oversmoothing on general geometries.

  12. Physics-Informed Neural PDE Solvers via Spatio-Temporal MeanFlow

    cs.LG 2026-05 unverdicted novelty 7.0

    Spatio-Temporal MeanFlow adapts MeanFlow to PDEs by replacing the generative velocity field with the physical operator and extending the integral constraint to the spatio-temporal domain, yielding a unified solver for...

  13. Controlling Transient Amplification Improves Long-horizon Rollouts

    cs.LG 2026-05 unverdicted novelty 7.0

    Commutativity regularization on Jacobians reduces transient error amplification in neural simulators, enabling stable rollouts over thousands of steps on physical and climate data.

  14. PerFlow: Physics-Embedded Rectified Flow for Efficient Reconstruction and Uncertainty Quantification of Spatiotemporal Dynamics

    cs.LG 2026-05 unverdicted novelty 7.0

    PerFlow embeds physics constraints into rectified flow sampling through guidance-free conditioning and constraint-preserving projections, achieving efficient sparse reconstruction and uncertainty quantification for sp...

  15. PODiff: Latent Diffusion in Proper Orthogonal Decomposition Space for Scientific Super-Resolution

    cs.LG 2026-05 unverdicted novelty 7.0

    PODiff performs conditional diffusion in a fixed, variance-ordered POD latent space to enable efficient probabilistic super-resolution of high-dimensional scientific fields with lower memory and better-calibrated unce...

  16. Quantitative Sobolev Approximation Bounds for Neural Operators with Empirical Validation on Burgers Equation

    cs.LG 2026-05 unverdicted novelty 7.0

    Neural operators approximate continuous operators from H^s to H^t with O(N^{-s/d}) error in H^t norm; FNOs on Burgers achieve H^1 errors to 10^{-7} and follow a power-law scaling with exponent ~1.4.

  17. Isotropic Fourier Neural Operators

    cs.LG 2026-05 unverdicted novelty 7.0

    Isotropic Fourier Neural Operators enforce spatial symmetries in Fourier layers, improving PDE-solving performance while reducing parameters by up to 16x in 2D and 96x in 3D.

  18. Online Safety Filter for Deformable Object Manipulation with Horizon Agnostic Neural Operators

    cs.RO 2026-05 unverdicted novelty 7.0

    A horizon-agnostic neural operator paired with a boundary control barrier function creates a real-time safety filter that raises safe trajectory rates by up to 22% on fluid manipulation tasks in simulation.

  19. An approach to encode divergence-free stress fields in neural approximations based on stress potentials

    cs.CE 2026-05 unverdicted novelty 7.0

    A physics-encoded Fourier neural operator (PeFNO) uses stress potentials to enforce divergence-free stress fields by architecture design, yielding better equilibrium satisfaction than physics-informed or physics-guide...

  20. Hybrid Fourier Neural Operator-Lattice Boltzmann Method

    physics.flu-dyn 2026-04 unverdicted novelty 7.0

    Hybrid FNO-LBM accelerates porous media flow convergence by up to 70% via neural initialization and stabilizes unsteady simulations through embedded FNO rollouts, allowing small models to match larger ones in accuracy.

  21. Robust Model-Based Iteration for Passive Gamma Emission Tomography

    math.NA 2026-04 unverdicted novelty 7.0

    A safeguarded hybrid of Levenberg-Marquardt and learned operators achieves equivalent reconstruction quality for PGET in roughly one-third the iterations, with architecture-dependent robustness.

  22. Learning Neural Operator Surrogates for the Black Hole Accretion Code

    astro-ph.HE 2026-04 unverdicted novelty 7.0

    Physics-informed Fourier neural operators recover plasmoid formation in sparse SRRMHD vortex data where data-only models fail, and transformer operators approximate AMR jet evolution, marking first reported uses in th...

  23. Droplet-LNO: Physics-Informed Laplace Neural Operators for Accurate Prediction of Droplet Spreading Dynamics on Complex Surfaces

    cs.LG 2026-04 unverdicted novelty 7.0

    PI-LNO is a physics-informed neural operator that uses Laplace transforms and fluid physics constraints to accurately and rapidly predict droplet spreading dynamics on complex surfaces.

  24. AI models of unstable flow exhibit hallucination

    physics.flu-dyn 2026-04 unverdicted novelty 7.0

    AI models of viscous fingering exhibit hallucinations from spectral bias; DeepFingers combines FNO and DeepONet with time-contrast conditioning to predict accurate finger dynamics while preserving mixing metrics.

  25. Faster by Design: Interactive Aerodynamics via Neural Surrogates Trained on Expert-Validated CFD

    cs.LG 2026-04 unverdicted novelty 7.0

    A graph-based neural operator trained on expert-validated race-car CFD data reaches accuracy levels usable for early-stage interactive aerodynamic design exploration.

  26. DeepRitzSplit Neural Operator for Phase-Field Models via Energy Splitting

    math.AP 2026-04 unverdicted novelty 7.0

    A DeepRitzSplit neural operator trained on energy-split variational forms enforces dissipation in phase-field models and outperforms data-driven training in generalization while running faster than Fourier spectral me...

  27. Neuroscience Inspired Graph Operators Towards Edge-Deployable Virtual Sensing for Irregular Geometries

    cs.LG 2026-04 unverdicted novelty 7.0

    VS-GNO delivers 0.71-1.04% reconstruction error at 15-24.5% spiking rates versus 0.4% for a non-spiking baseline in sparse-to-dense virtual sensing.

  28. G-PARC: Graph-Physics Aware Recurrent Convolutional Neural Networks for Spatiotemporal Dynamics on Unstructured Meshes

    cs.LG 2026-04 unverdicted novelty 7.0

    G-PARC embeds analytically computed differential operators via moving least squares on graphs into recurrent networks, achieving higher accuracy with 2-3x fewer parameters than prior graph PADL methods on nonlinear be...

  29. DiLO: Decoupling Generative Priors and Neural Operators via Diffusion Latent Optimization for Inverse Problems

    math.NA 2026-04 unverdicted novelty 7.0

    DiLO turns diffusion sampling into deterministic latent optimization to satisfy the manifold consistency requirement for neural operators in inverse problem solving.

  30. Learning on the Temporal Tangent Bundle for Physics-Informed Neural Networks

    math.NA 2026-04 unverdicted novelty 7.0

    Parameterizing the temporal derivative in PINNs and reconstructing via Volterra integral yields 100-200x lower errors on advection, Burgers, and Klein-Gordon equations while proving equivalence to the original PDE.

  31. Kathleen: Oscillator-Based Byte-Level Text Classification Without Tokenization or Attention

    cs.CL 2026-04 unverdicted novelty 7.0

    Kathleen uses recurrent oscillator banks, an efficient wavetable encoder, and phase harmonics to classify text at the byte level with high accuracy and low parameter count.

  32. Kathleen: Oscillator-Based Byte-Level Text Classification Without Tokenization or Attention

    cs.CL 2026-04 unverdicted novelty 7.0

    Kathleen performs byte-level text classification via recurrent oscillator banks, FFT wavetable encoding, and phase harmonics, matching pretrained baselines on standard benchmarks with 36% fewer parameters.

  33. SPAMoE: Spectrum-Aware Hybrid Operator Framework for Full-Waveform Inversion

    cs.LG 2026-04 unverdicted novelty 7.0

    SPAMoE reduces average MAE by 44.4% on OpenFWI datasets for full-waveform inversion via a spectral-preserving DINO encoder and dynamic frequency-band routing to specialized neural operators.

  34. Generative modeling of granular flow on inclined planes using conditional flow matching

    cs.CE 2026-04 unverdicted novelty 7.0

    A conditional flow matching model trained on DEM simulations reconstructs granular flow velocity fields from as little as 11-16% sparse boundary data, outperforming deterministic CNN baselines while providing uncertai...

  35. A Unified Multiscale Auxiliary PINN Framework for Generalized Phonon Transport

    cond-mat.mes-hall 2026-03 unverdicted novelty 7.0

    MTNet is a new auxiliary PINN framework that solves the generalized equation of phonon radiative transfer by converting it to a differential system, capturing multiscale phonon transport in nanostructures beyond stand...

  36. SLE-FNO: Single-Layer Extensions for Task-Agnostic Continual Learning in Fourier Neural Operators

    cs.LG 2026-03 unverdicted novelty 7.0

    SLE-FNO achieves zero forgetting and strong plasticity-stability balance in continual learning for FNO surrogate models of pulsatile blood flow by adding minimal single-layer extensions across four out-of-distribution tasks.

  37. Tackling multiphysics problems via finite element-guided physics-informed operator learning

    cs.LG 2026-03 unverdicted novelty 7.0

    A finite element-guided physics-informed operator learning framework learns solution operators for coupled multiphysics PDEs, enabling discretization-independent predictions on arbitrary domains without labeled data.

  38. Symbolic recovery of PDEs from measurement data

    cs.LG 2026-02 unverdicted novelty 7.0

    Symbolic rational-function networks recover an admissible PDE from noiseless complete measurements and select the regularization-minimizing parameterization within the architecture.

  39. Fluids You Can Trust: Property-Preserving Operator Learning for Incompressible Flows

    physics.flu-dyn 2026-02 conditional novelty 7.0

    A kernel operator learning framework constructs property-preserving bases so that predicted incompressible velocity fields satisfy divergence-free and periodicity conditions exactly, delivering up to six orders lower ...

  40. Latent Generative Solvers for Generalizable Long-Term Physics Simulation

    cs.AI 2026-02 unverdicted novelty 7.0

    LGS pretrained on 2.5M trajectories across 16 systems matches deterministic baselines at one step and halves 20-step error while using far less compute and adapting to held-out higher-resolution flows.

  41. DuFal: Dual-Frequency-Aware Learning for High-Fidelity Extremely Sparse-view CBCT Reconstruction

    cs.CV 2026-01 unverdicted novelty 7.0

    DuFal combines global and local high-frequency Fourier neural operators with cross-attention fusion to recover fine anatomical structures in extremely sparse-view CBCT, outperforming prior methods on LUNA16 and ToothF...

  42. CompNO: A Novel Foundation Model approach for solving Partial Differential Equations

    cs.LG 2026-01 unverdicted novelty 7.0

    CompNO composes specialized Fourier neural operator blocks for fundamental differential operators into task-specific solvers that achieve lower L2 error than baselines on linear parametric PDEs and remain competitive ...

  43. An Energy Stable Approach for Learning Derivative Operators from Noisy Data for Maxwells Equations

    math.NA 2026-01 unverdicted novelty 7.0

    SP-ADMM learns energy-stable derivative stencils for Maxwell equations from noisy data by enforcing skew-adjointness through reduced parameterization of periodic convolution stencils.

  44. A Wachspress-based transfinite formulation for exactly enforcing Dirichlet boundary conditions on convex polygonal domains in physics-informed neural networks

    math.NA 2026-01 unverdicted novelty 7.0

    Wachspress coordinates enable a transfinite interpolant that exactly enforces Dirichlet boundary conditions in PINNs on convex polygons by providing a smooth lift of boundary data into the domain interior.

  45. Kinetic-Mamba: Mamba-Assisted Predictions of Stiff Chemical Kinetics

    cs.LG 2025-12 unverdicted novelty 7.0

    Mamba-based neural operators predict stiff chemical kinetics evolution with high fidelity from initial states on Syngas and GRI-Mech 3.0 mechanisms.

  46. Walsh-Hadamard Neural Operators for Solving PDEs with Discontinuous Coefficients

    physics.comp-ph 2025-11 conditional novelty 7.0

    WHNO using Walsh-Hadamard transforms outperforms Fourier Neural Operators on PDEs with discontinuous coefficients, and optimal WHNO-FNO ensembles cut mean-squared error by 35-40 percent.

  47. PVD-ONet: A Multi-scale Neural Operator Method for Singularly Perturbed Boundary Layer Problems

    cs.LG 2025-07 unverdicted novelty 7.0

    PVD-ONet combines multi-network DeepONet modules with Prandtl and Van Dyke matching conditions to map initial data to solution operators for families of singularly perturbed boundary-layer problems and to infer scalin...

  48. Beyond Blur: A Fluid Perspective on Generative Diffusion Models

    cs.GR 2025-06 unverdicted novelty 7.0

    Proposes an advection-diffusion PDE corruption process with stochastic velocity fields and Lattice Boltzmann solver for diffusion models, generalizing prior PDE methods.

  49. SPDEBench: An Extensive Benchmark for Learning Stochastic PDEs

    cs.LG 2025-05 conditional novelty 7.0

    SPDEBench supplies standardized datasets, baselines, and metrics for machine-learning approximation of stochastic PDEs and finds that SPDE-aware architectures outperform generic operator learners.

  50. SPDEBench: An Extensive Benchmark for Learning Stochastic PDEs

    cs.LG 2025-05 unverdicted novelty 7.0

    SPDEBench is the first unified benchmark providing ready-to-use datasets for regular and singular SPDEs, ML operator-learning baselines, and evaluations showing SPDE-aware models outperform generic ones on accuracy, r...

  51. Non-markovian neural quantum propagator and its application to the simulation of ultrafast nonlinear spectra

    physics.chem-ph 2024-08 unverdicted novelty 7.0

    A machine learning model called neural quantum propagator is introduced to efficiently solve non-Markovian quantum dynamics described by HEOM and applied to simulate spectra of the FMO complex.

  52. Neural Green's Operators for Parametric Partial Differential Equations

    cs.LG 2024-06 unverdicted novelty 7.0

    Neural Green's Operators approximate the Green's function and its parametric dependence for linear PDEs via neural networks, preserving linearity and enabling use in time-stepping, nonlinear solvers, and preconditioning.

  53. Discontinuous Galerkin Neural Operator for Pathology Defocus Deblurring

    eess.IV 2026-05 unverdicted novelty 6.0

    DGNO parameterizes integral kernels with discontinuous Galerkin elements for heterogeneous defocus deblurring in pathology images and reports superior performance over prior methods.

  54. Therm-FM: Foundation Model is ALL YOU NEED for 3D-ICs Thermal Simulation

    cs.CE 2026-05 unverdicted novelty 6.0

    Therm-FM adapts pretrained diffusion PDE foundation models to 3D-IC thermal simulation with multi-fidelity adaptation, reporting up to 10.6x mean error reduction and strong cross-design performance using under 20% of ...

  55. IKNO: Infinite-order Kernel Neural Operators

    cs.LG 2026-05 unverdicted novelty 6.0

    IKNO replaces first-order kernel integrals in neural operators with infinite-order versions that have efficient closed-form approximations and reports SOTA accuracy on time-dependent and time-independent benchmarks.

  56. Data-Efficient Neural Operator Training via Physics-Based Active Learning

    cs.LG 2026-05 unverdicted novelty 6.0

    Physics-based active learning using PDE residuals improves data efficiency for neural operator training on Burgers and Navier-Stokes equations while adding a physics inductive bias.

  57. Learning to Think in Physics: Breaking Shortcut Learning in Scientific Diffusion via Representation Alignment

    cs.LG 2026-05 unverdicted novelty 6.0

    REPA-P aligns intermediate representations in diffusion models with physical states using first-principles PDE residuals to accelerate convergence and boost out-of-distribution robustness on PDE tasks.

  58. When can a neural operator replace a coarse solve? Architectural principles for two-level preconditioning

    math.NA 2026-05 unverdicted novelty 6.0

    The Neural Green's Operator matches exact coarse-solve iteration counts in two-level preconditioners for diffusion and advection-diffusion problems when inputs are integrated against the output basis.

  59. Smooth Piecewise Cutting for Neural Operator to Handle Discontinuities and Sharp Transitions

    cs.LG 2026-05 conditional novelty 6.0

    Cut-DeepONet uses a lifting strategy and an auxiliary network to predict discontinuity locations, enabling a neural operator to learn smooth components in partitioned regions and outperforming prior methods on benchma...

  60. Neural Operators for Design-Space Surrogate Modeling of Tendon-Actuated Continuum Robots

    cs.RO 2026-05 unverdicted novelty 6.0

    Develops four neural operator architectures to create generalizable surrogate models for tendon-driven continuum robot configurations across varying designs using simulation data.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · cited by 169 Pith papers · 1 internal anchor

  1. [1]

    Solving ill-posed inverse problems using iterative deep neural networks,

    doi: 10.1088/1361-6420/aa9581. URL https: //doi.org/10.1088%2F1361-6420%2Faa9581. Leah Bar and Nir Sochen. Unsupervised deep learning algorithm for pde-based forward and inverse problems. arXiv preprint arXiv:1904.05417,

    work page doi:10.1088/1361-6420/aa9581 1904

  2. [2]

    doi: 10.1214/13-sts421

    ISSN 0883-4237. doi: 10.1214/13-sts421. URL http://dx.doi.org/10.1214/13-STS421. Ronald A. DeV ore.Chapter 3: The Theoretical Foundation of Reduced Basis Methods

    work page doi:10.1214/13-sts421

  3. [3]

    DeVore , title =

    doi: 10.1137/1.9781611974829.ch3. URL https://epubs.siam.org/doi/abs/10.1137/ 1.9781611974829.ch3. Weinan E and Bing Yu. The deep ritz method: A deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics , 3

    work page doi:10.1137/1.9781611974829.ch3

  4. [4]

    The Deep Ritz Method: A Deep Learning-Based Numerical Algo- rithm for Solving Variational Problems,

    ISSN 2194-6701. doi: 10.1007/s40304-018-0127-z. Yuwei Fan, Cindy Orozco Bohorquez, and Lexing Ying. Bcr-net: A neural network based on the nonstandard wavelet form. Journal of Computational Physics, 384:1–15, 2019a. Yuwei Fan, Lin Lin, Lexing Ying, and Leonardo Zepeda-N´unez. A multiscale neural network based on hierarchical matrices. Multiscale Modeling ...

    work page doi:10.1007/s40304-018-0127-z

  5. [5]

    Deep residual learning for image recog- nition

    10 Published as a conference paper at ICLR 2021 Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 770–778,

    work page 2021

  6. [6]

    Meshfreeflownet: A physics-constrained deep continuous space-time super-resolution framework

    Chiyu Max Jiang, Soheil Esmaeilzadeh, Kamyar Azizzadenesheli, Karthik Kashinath, Mustafa Mustafa, Hamdi A Tchelepi, Philip Marcus, Anima Anandkumar, et al. Meshfreeflownet: A physics-constrained deep continuous space-time super-resolution framework. arXiv preprint arXiv:2005.01463,

    work page arXiv 2005

  7. [7]

    Enforcing physical constraints in cnns through differen- tiable pde layer

    Karthik Kashinath, Philip Marcus, et al. Enforcing physical constraints in cnns through differen- tiable pde layer. In ICLR 2020 Workshop on Integration of Deep Neural Models and Differential Equations,

    work page 2020

  8. [8]

    Solving parametric PDE problems with artificial neural networks

    Yuehaw Khoo, Jianfeng Lu, and Lexing Ying. Solving parametric PDE problems with artificial neural networks. arXiv preprint arXiv:1707.03351,

    work page arXiv

  9. [9]

    Proceedings of the National Academy of Sciences118(21), 2101784118 (2021) https://doi.org/10

    Dmitrii Kochkov, Jamie A Smith, Ayya Alieva, Qing Wang, Michael P Brenner, and Stephan Hoyer. Machine learning accelerated computational fluid dynamics. arXiv preprint arXiv:2102.01010,

    work page arXiv

  10. [10]

    Neural Operator: Graph Kernel Network for Partial Differential Equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, An- drew Stuart, and Anima Anandkumar. Multipole graph neural operator for parametric partial differential equations, 2020a. Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, An- drew Stuart, and Anima Anandkumar. Neural operator: ...

    work page internal anchor Pith review arXiv 2003

  11. [11]

    The random feature model for input-output maps between banach spaces

    NH Nelsen and AM Stuart. The random feature model for input-output maps between banach spaces. arXiv preprint arXiv:2005.10224,

    work page arXiv 2005

  12. [12]

    Sitzmann, J

    Vincent Sitzmann, Julien NP Martel, Alexander W Bergman, David B Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions. arXiv preprint arXiv:2006.09661,

    work page arXiv 2006

  13. [13]

    Eikonet: Solving the eikonal equation with deep neural networks

    11 Published as a conference paper at ICLR 2021 Jonathan D Smith, Kamyar Azizzadenesheli, and Zachary E Ross. Eikonet: Solving the eikonal equation with deep neural networks. arXiv preprint arXiv:2004.00361,

    work page arXiv 2021

  14. [14]

    doi: 10.1016/j.jcp.2018.04.018

    ISSN 0021-9991. doi: https://doi.org/10.1016/j.jcp.2018.04.018. URL http://www. sciencedirect.com/science/article/pii/S0021999118302341. 12 Published as a conference paper at ICLR 2021 A A PPENDIX A.1 T ABLE OF NOTATIONS A table of notations is given in Table

    work page doi:10.1016/j.jcp.2018.04.018 2018

  15. [15]

    The y-axis is the spectrum; the x-axis is the wavenumber|k| =k1 +k2

    13 Published as a conference paper at ICLR 2021 The spectral decay of the Navier-stokes equation data we used in section 5.3. The y-axis is the spectrum; the x-axis is the wavenumber|k| =k1 +k2. Figure 4: Spectral Decay of Navier-Stokes equations The error of truncation in one single Fourier layer without applying the linear transformR. The y-axis is the ...

    work page 2021

  16. [16]

    The initial condition u0(x) is generated according to u0∼ µ where µ =N (0, 625(−∆ + 25I)−2) with periodic boundary conditions

    = u0(x), x ∈ (0, 1). The initial condition u0(x) is generated according to u0∼ µ where µ =N (0, 625(−∆ + 25I)−2) with periodic boundary conditions. We set the viscosity to ν = 0.1 and solve the equation using a split step method where the heat equation part is solved exactly in Fourier space then the non-linear part is advanced, again in Fourier space, us...

    work page 2021

  17. [17]

    Table 3: Benchmarks on 1-d Burgers’ equation Networks s = 256 s = 512 s = 1024 s = 2048 s = 4096 s = 8192 NN 0.4714 0 .4561 0 .4803 0 .4645 0 .4779 0 .4452 GCN 0.3999 0 .4138 0 .4176 0 .4157 0 .4191 0 .4198 FCN 0.0958 0 .1407 0 .1877 0 .2313 0 .2855 0 .3238 PCANN 0.0398 0 .0395 0 .0391 0 .0383 0 .0392 0 .0393 GNO 0.0555 0 .0594 0 .0651 0 .0663 0 .0666 0 ....

    work page 2048

  18. [18]

    It can be seen that the result using Fourier neural operator as a surrogate is as good as the result of the traditional solver. 15 Published as a conference paper at ICLR 2021 The top left panel shows the true initial vorticity while bottom left panel shows the true observed vorticity at T = 50 with black dots indicating the locations of the observation p...

    work page 2021