Pith. sign in

REVIEW 3 major objections 2 minor 67 references

A grid-spacing-aware DeepONet hybrid accelerates the pressure Poisson equation on non-uniform Cartesian grids and transfers, with fixed weights, to immersed-boundary flows after training only on fabricated linear systems.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 14:11 UTC pith:XAWNRSU2

load-bearing objection Wrong full text is attached (VIRSO, not Bai et al.); 2604.01800 is abstract-only, so the fabricated-system PPE claim cannot be audited. the 3 major comments →

arxiv 2604.01800 v3 pith:XAWNRSU2 submitted 2026-04-02 physics.flu-dyn

Deep learning accelerated solutions of incompressible Navier-Stokes equations on non-uniform Cartesian grids

Heming Bai , Dong Zhang , Shengze Cai , Xin Bian This is my paper
classification physics.flu-dyn
keywords pressure Poisson equationnon-uniform Cartesian gridsDeepONetU-Net branchmulti-level distance vector mapsimmersed boundary methodhybrid iterative solversincompressible Navier-Stokes
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Incompressible flow solvers on non-uniform Cartesian grids spend most of their time on the pressure Poisson equation, whose linear systems become especially hard when spacing varies. This paper claims that a hybrid solver—classical iterative methods guided by a Deep Operator Network whose U-Net branch is fed multi-level distance-vector maps of the local grid spacing—can solve those systems far faster than a preconditioned conjugate gradient method alone, and faster than the same network without the spacing maps. Because the network is trained only on fabricated linear systems rather than on flow snapshots, the same fixed weights are said to remain accurate when the hybrid is embedded in a fractional-step scheme and coupled to a decoupled immersed-boundary projection, covering diverse solid obstacles without retraining. The practical stake is a deployable route to high-resolution, geometry-rich CFD that still respects classical mass conservation while cutting the dominant linear-algebra cost.

Core claim

A DeepONet with a U-Net branch that fuses multi-level discrete grid-spacing maps into every hierarchical feature map, trained solely on fabricated linear systems, accelerates PPE solves on non-uniform Cartesian grids and, with unchanged weights, extends via a decoupled immersed-boundary projection to flows past diverse solid obstacles, outperforming both standalone preconditioned conjugate gradient and a standard-convolution counterpart.

What carries the argument

Multi-level distance vector maps: at each U-Net hierarchy level the discrete local grid spacings are computed and explicitly fused into the feature maps before convolution, giving the operator explicit awareness of spatially varying resolution so that the same fixed network can act as a geometry-agnostic accelerator for the PPE (and for immersed-boundary PPE variants).

Load-bearing premise

Training only on made-up linear systems, not on real flow data, is enough for the operator to stay accurate and stable inside full time-marching Navier–Stokes runs across many immersed shapes without retraining or breaking mass conservation.

What would settle it

Embed the fixed-weight hybrid in a fractional-step solver, run several immersed-boundary benchmarks with different obstacle shapes, and check whether PPE residual, velocity divergence, and temporal accuracy remain at least as good as a well-tuned PCG baseline while wall-clock PPE time drops; any large residual growth, mass-loss, or need for geometry-specific fine-tuning would falsify the central claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • PPE-dominated fractional-step codes on stretched Cartesian meshes can swap in the hybrid without rewriting the outer time scheme.
  • Immersed-boundary simulations of many obstacle shapes can reuse one set of network weights after a single offline training on synthetic linear systems.
  • Grid-spacing fusion becomes a reusable design pattern for operator networks that must act on non-uniform discretizations.
  • Real-world CFD wall-clock budgets can shift from linear algebra toward higher resolution or more design cases once the PPE bottleneck is cut.

Where Pith is reading between the lines

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

  • If fabricated-system training really transfers, the same idea may accelerate other elliptic subproblems in multiphase or low-Mach solvers that also face non-uniform stencils.
  • The multi-level distance maps are essentially a cheap inductive bias for mesh anisotropy; similar maps could be tried inside pure learned surrogates, not only hybrid iterative solvers.
  • A natural next stress test is whether the fixed weights survive strongly unsteady free-surface or moving-body cases where the PPE spectrum changes every step.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 2 minor

Summary. The submission’s title and abstract claim an extended hybrid DeepONet framework (U-Net branch with multi-level distance-vector maps fused before convolutions) that accelerates the pressure Poisson equation on non-uniform Cartesian grids, is trained only on fabricated linear systems, extends via a decoupled immersed-boundary projection method to diverse solid geometries with fixed weights, and significantly outperforms preconditioned conjugate gradient and a standard-convolution hybrid. The body of the manuscript provided for review is an entirely different work (VIRSO: a spatial–spectral graph neural operator for sparse boundary-to-interior multiphysics reconstruction, with Jetson/H200 deployment results and nuclear thermal-hydraulic benchmarks). No methods, equations, fabricated-system construction, hybrid residual coupling, conservation checks, or PPE/IB flow benchmarks for the claimed Navier–Stokes contribution appear in the supplied full text.

Significance. If the abstract’s claims were substantiated—grid-spacing-aware operator learning of non-uniform PPE, geometry-agnostic generalization from fabricated systems alone, and stable embedding in fractional-step/IB time marching with clear gains over PCG—the work would be a meaningful contribution to hybrid CFD solvers and operator learning for variable-resolution grids. That significance cannot be assessed from the materials provided, because the manuscript body does not develop or evaluate those claims.

major comments (3)
  1. Title/abstract vs. full text mismatch: the abstract and paper_id describe a DeepONet–U-Net hybrid for non-uniform-grid PPE and immersed-boundary NS; the full manuscript is VIRSO (graph spectral–spatial operator, V-KNN, edge deployment on Jetson Orin Nano, LDC/PWR/heat-exchanger sensing). No section develops multi-level distance vector maps, fabricated PPE systems, fractional-step coupling, or PCG baselines. The central claims of 2604.01800 are therefore not present in the document under review and cannot be audited.
  2. Load-bearing generalization premise (abstract): training exclusively on fabricated linear systems with fixed weights across diverse immersed geometries is the hinge of the claimed contribution. The supplied manuscript contains no definition of those systems, no spectrum/conditioning analysis relative to real non-uniform PPE operators under IB projection, and no coupled fractional-step residual, mass-conservation, or temporal-accuracy results. This premise remains unverifiable.
  3. Benchmark claims (abstract: “significantly outperforms standalone preconditioned conjugate gradient… and its standard convolution counterpart”): no tables, residual histories, iteration counts, wall-clock timings, or error metrics for PPE or full NS/IB runs appear for the claimed method. Without those, the performance claim cannot be evaluated.
minor comments (2)
  1. If the intended submission is the VIRSO manuscript, the title, abstract, paper_id, and primary category must be corrected and the package resubmitted as a coherent unit; the present package is not reviewable as either paper.
  2. Abstract-only claims for 2604.01800 use strong language (“effortlessly,” “exceptional potential”) that would need quantitative support (error tables, conservation diagnostics, geometry suite) in any future correct submission.

Circularity Check

0 steps flagged

No circular derivation: the PPE/DeepONet hybrid claim is empirical (train operator, measure solver metrics); the supplied full text is a different paper (VIRSO) and likewise reports measured operator accuracy, not results forced by definition or self-citation.

full rationale

Circularity requires a load-bearing step that reduces by construction to its own inputs (self-definition, fitted quantity renamed as prediction, or uniqueness forced only by overlapping-author citation). The target abstract (2604.01800) states an empirical hybrid solver: a DeepONet with U-Net branch and multi-level distance-vector maps is trained on fabricated linear systems and then timed/accuracy-checked against PCG and a plain-convolution hybrid on non-uniform PPE and immersed-boundary flows. That is train-then-measure, not a tautology; whether fabricated systems span real PPE spectra is a generalization risk, not circularity. The CACHEABLE full manuscript is a different work (VIRSO, sparse-boundary multiphysics reconstruction). VIRSO likewise learns G from Fluent data, reports relative L2 on held-out splits, and compares energy/latency to Geo-FNO, NOMAD, and GNO; ablations and V-KNN are empirical, not definitional identities. Self-citations (e.g. Sp2GNO) supply architectural prior art but do not force the reported errors or hardware numbers. No equation equates a fitted parameter to a claimed prediction; no uniqueness theorem from the same authors forbids alternatives. Score 0; steps empty.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 1 invented entities

Abstract-only review of 2604.01800. Load-bearing premises are methodological choices typical of hybrid CFD–ML papers: fractional-step PPE bottleneck, usefulness of non-uniform Cartesian grids, validity of training on synthetic linear systems, and transfer to immersed-boundary coupling with frozen weights. No free parameters or invented physical entities can be extracted beyond architecture hyperparameters that would appear in a full methods section.

free parameters (1)
  • Network and training hyperparameters (depth/width of U-Net branch, number of distance-vector levels, fabricated-system d
    Not specified in the abstract; any quantitative speedup claim will depend on these choices once the full paper is available.
axioms (3)
  • domain assumption The pressure Poisson equation on non-uniform Cartesian grids is the primary computational bottleneck of the fractional-step method for incompressible Navier–Stokes.
    Stated as motivation in the abstract; standard in CFD but problem-dependent (preconditioner quality, Reynolds number, grid stretching).
  • ad hoc to paper Explicit multi-level discrete grid-spacing maps fused into U-Net feature maps before convolution are sufficient to capture spatially varying resolution for operator learning of the PPE.
    Core architectural hypothesis of the work; not a standard theorem.
  • ad hoc to paper Training only on fabricated linear systems yields a model that generalizes across diverse immersed obstacle geometries with fixed weights inside a decoupled immersed-boundary projection method.
    Central generalization claim; load-bearing and not independently established in the abstract.
invented entities (1)
  • Multi-level distance vector map construction strategy for U-Net hierarchical levels no independent evidence
    purpose: Inject discrete grid-spacing information at each U-Net level so convolutions respect non-uniform Cartesian resolution.
    Named methodological construct of the paper; independent evidence would be ablations vs. standard convolution (claimed but not shown here).

pith-pipeline@v1.1.0-grok45 · 42199 in / 2876 out tokens · 30074 ms · 2026-07-13T14:11:39.792443+00:00 · methodology

0 comments
read the original abstract

In incompressible flow simulations, non-uniform grids efficiently capture localized flow features; however, their spatially varying resolutions severely exacerbate computational complexity. The pressure Poisson equation (PPE) formulated on these grids yields highly complex linear systems, forming the primary computational bottleneck in fractional step method. To address this, we develop an extended hybrid framework tailored for non-uniform Cartesian grids, integrating deep learning with classical iterative solvers to accelerate PPE solutions. Specifically, the framework employs a deep operator network with a U-Net-based branch network. To effectively capture spatially varying resolutions, we propose a multi-level distance vector map construction strategy that computes discrete grid-spacing information corresponding to each hierarchical level of the U-Net. This grid-spacing information is explicitly fused into feature maps prior to convolution operations. Empowered by this grid-spacing-aware architecture, the framework seamlessly extends to simulate flows interacting with solid structures using a decoupled immersed boundary projection method. By training exclusively on fabricated linear systems rather than conventional flow-dependent datasets, the model generalizes effortlessly across diverse immersed obstacle geometries with fixed network weights. Benchmark results demonstrate that the framework significantly outperforms standalone preconditioned conjugate gradient methods and its standard convolution counterpart, underscoring its exceptional potential for real-world computational fluid dynamics applications.

discussion (0)

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

Reference graph

Works this paper leans on

67 extracted references · 3 linked inside Pith

  1. [1]

    From proxies to fields: Spatiotemporal reconstruction of global radiation from sparse sensor sequences.����� �������� ����������������, 2025

    Kazuma Kobayashi, Roy, et al. From proxies to fields: Spatiotemporal reconstruction of global radiation from sparse sensor sequences.����� �������� ����������������, 2025

    2025

  2. [2]

    Sensor degradation in nuclear reactor pressure vessels: the overlooked factor in remaining useful life prediction.��� ��������� �����������, 8(1):71, 2024

    Raisa Bentay Hossain, Kazuma Kobayashi, and Syed Bahauddin Alam. Sensor degradation in nuclear reactor pressure vessels: the overlooked factor in remaining useful life prediction.��� ��������� �����������, 8(1):71, 2024

    2024

  3. [3]

    Review of nuclear microreactors: Status, potentialities and challenges.�������� �� ������� ������, 138:103822, 2021

    Raffaella Testoni, Andrea Bersano, and Stefano Segantin. Review of nuclear microreactors: Status, potentialities and challenges.�������� �� ������� ������, 138:103822, 2021

    2021

  4. [4]

    The role of advanced nuclear reactors and fuel cycles in a future energy system–clean.���� �����, page pgae030, 2024

    Kasia Kornecki and Catherine F Wise. The role of advanced nuclear reactors and fuel cycles in a future energy system–clean.���� �����, page pgae030, 2024

    2024

  5. [5]

    When network architecture meets physics: Deep operator learning for coupled multiphysics.����� �������� ����������������, 2025

    Kazuma Kobayashi, Jaewan Park, Qibang Liu, Seid Koric, Diab Abueidda, and Syed Bahauddin Alam. When network architecture meets physics: Deep operator learning for coupled multiphysics.����� �������� ����������������, 2025. 30 Computational Sensing of Inaccessible Physical Fields

    2025

  6. [6]

    Bridging sequential deep operator network and video diffusion: Residual refinement of spatio-temporal pde solutions.����� �������� ����������������, 2025

    Jaewan Park, Farid Ahmed, Kazuma Kobayashi, Seid Koric, Syed Bahauddin Alam, Iwona Jasiuk, and Diab Abueidda. Bridging sequential deep operator network and video diffusion: Residual refinement of spatio-temporal pde solutions.����� �������� ����������������, 2025

    2025

  7. [7]

    Improved generalization with deep neural operators for engineering systems: Path towards digital twin.����������� ������������ �� ��������� ������������, 131:107844, 2024

    Kazuma Kobayashi, James Daniell, and Syed Bahauddin Alam. Improved generalization with deep neural operators for engineering systems: Path towards digital twin.����������� ������������ �� ��������� ������������, 131:107844, 2024

    2024

  8. [8]

    Deep neural operator-driven real-time inference to enable digital twin solutions for nuclear energy systems.��������� �������, 14(1):2101, 2024

    Kazuma Kobayashi and Syed Bahauddin Alam. Deep neural operator-driven real-time inference to enable digital twin solutions for nuclear energy systems.��������� �������, 14(1):2101, 2024

    2024

  9. [9]

    Hasekamp.������ �������� ��� ������� ������ �������� ��� ����� ����������� �� ������� ��� ����� ������ �������� � ������, pages 279–329

    Antonio Di Noia and Otto P. Hasekamp.������ �������� ��� ������� ������ �������� ��� ����� ����������� �� ������� ��� ����� ������ �������� � ������, pages 279–329. Springer International Publishing, Cham, 2018

    2018

  10. [10]

    Schiller and Matthias A

    Julian D. Schiller and Matthias A. Müller. Nonlinear moving horizon estimation for robust state and parameter estimation.����������, 185:112790, 2026

    2026

  11. [11]

    Raissi, P

    M. Raissi, P. Perdikaris, and G.E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.������� �� ������������� �������, 378:686–707, 2019

    2019

  12. [12]

    Brunton, Joshua L

    Steven L. Brunton, Joshua L. Proctor, and J. Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems.����������� �� ��� �������� ������� �� ��������, 113(15):3932–3937, March 2016

    2016

  13. [13]

    Syed Bahauddin Alam, Bader Almutairi, Tuhfatur Ridwan, Dinesh Kumar, Cameron S Goodwin, Kirk D Atkinson, and Geoffrey T Parks. Neutronic investigation of alternative & composite burnable poisons for the soluble-boron- free and long life civil marine small modular reactor cores.��������� �������, 9(1):19591, 2019

    2019

  14. [14]

    Small modular reactor core design for civil marine propulsion using micro-heterogeneous duplex fuel

    Syed Bahauddin Alam, Dinesh Kumar, Bader Almutairi, Palash Kumar Bhowmik, Cameron Goodwin, and Geoffrey T Parks. Small modular reactor core design for civil marine propulsion using micro-heterogeneous duplex fuel. part i: Assembly-level analysis.������� ����������� ��� ������, 346:157–175, 2019

    2019

  15. [15]

    Small modular reactor core design for civil marine propulsion using micro-heterogeneous duplex fuel

    Syed Bahauddin Alam, Tuhfatur Ridwan, Dinesh Kumar, Bader Almutairi, Cameron Goodwin, and Geoffrey T Parks. Small modular reactor core design for civil marine propulsion using micro-heterogeneous duplex fuel. part ii: Whole-core analysis.������� ����������� ��� ������, 346:176–191, 2019

    2019

  16. [16]

    Digital twin-centered hybrid data-driven multi-stage deep learning framework for enhanced nuclear reactor power prediction.������ ��� ��, 19:100450, 2025

    James Daniell, Kazuma Kobayashi, Ayodeji Alajo, and Syed Bahauddin Alam. Digital twin-centered hybrid data-driven multi-stage deep learning framework for enhanced nuclear reactor power prediction.������ ��� ��, 19:100450, 2025

    2025

  17. [17]

    Physics-regularized neural networks for predictive modeling of silicon carbide swelling with limited experimental data.��������� �������, 14(1):30666, 2024

    Kazuma Kobayashi and Syed Bahauddin Alam. Physics-regularized neural networks for predictive modeling of silicon carbide swelling with limited experimental data.��������� �������, 14(1):30666, 2024

    2024

  18. [18]

    Advances in digital twins and AI/ML for condition monitoring in nuclear applications.������ ������ ����, 14(1716514), February 2026

    Hector Mendoza, Vaibhav Yadav, Syed Bahauddin Alam, Doug Eskins, and Raj Iyengar. Advances in digital twins and AI/ML for condition monitoring in nuclear applications.������ ������ ����, 14(1716514), February 2026

    2026

  19. [19]

    Shaping the future of nuclear reactors with digital twins: Current developments and perspectives.������� ������, 402:126922, 2025

    Qingyu Huang, Wei Zeng, Jia Liu, Zhuo Zhang, Jian Deng, Zhifang Qiu, Le Xu, Zonglan Wei, Qi Lu, Lanxin Gong, Chunsen Shi, and Xianping Zhong. Shaping the future of nuclear reactors with digital twins: Current developments and perspectives.������� ������, 402:126922, 2025

    2025

  20. [20]

    Virtual sensing-enabled digital twin framework for real-time monitoring of nuclear systems leveraging deep neural operators.��� ��������� �����������, 9(1), Mar 2025

    Raisa Hossain, Farid Ahmed, Kazuma Kobayashi, Seid Koric, Diab Abueidda, and Syed Bahauddin Alam. Virtual sensing-enabled digital twin framework for real-time monitoring of nuclear systems leveraging deep neural operators.��� ��������� �����������, 9(1), Mar 2025

    2025

  21. [21]

    Kazuma Kobayashi and Syed Bahauddin Alam. Explainable, interpretable, and trustworthy AI for an intelligent digital twin: A case study on remaining useful life.����������� ������������ �� ��������� ������������, 129:107620, 2024

    2024

  22. [22]

    AI-driven non-intrusive uncertainty quantification of advanced nuclear fuels for digital twin-enabling technology.�������� �� ������� ������, 172:105177, 2024

    Kazuma Kobayashi, Dinesh Kumar, and Syed Bahauddin Alam. AI-driven non-intrusive uncertainty quantification of advanced nuclear fuels for digital twin-enabling technology.�������� �� ������� ������, 172:105177, 2024

    2024

  23. [23]

    Accurate medium-range global weather forecasting with 3D neural networks.������, 619(7970):533–538, July 2023

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Accurate medium-range global weather forecasting with 3D neural networks.������, 619(7970):533–538, July 2023

    2023

  24. [24]

    Physics-informed neural networks (pinns) for fluid mechanics: A review, 2021

    Shengze Cai, Zhiping Mao, Zhicheng Wang, Minglang Yin, and George Em Karniadakis. Physics-informed neural networks (pinns) for fluid mechanics: A review, 2021

    2021

  25. [25]

    Prashant K. Jha. From theory to application: A practical introduction to neural operators in scientific computing, 2025. 31 Computational Sensing of Inaccessible Physical Fields

    2025

  26. [26]

    Neural operator: Learning maps between function spaces with applications to pdes.������� �� ������� �������� ��������, 24(89):1–97, 2023

    Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes.������� �� ������� �������� ��������, 24(89):1–97, 2023

    2023

  27. [27]

    Adversarial vulnerabilities in neural operator digital twins: Gradient-free attacks on nuclear thermal-hydraulic surrogates.����� �������� ����������������, 2026

    Samrendra Roy et al. Adversarial vulnerabilities in neural operator digital twins: Gradient-free attacks on nuclear thermal-hydraulic surrogates.����� �������� ����������������, 2026

    2026

  28. [28]

    Sathujoda, Krzysztof Sawicki, Kanishk Gandhi, Angelica I

    Yuan Wang, Surya T. Sathujoda, Krzysztof Sawicki, Kanishk Gandhi, Angelica I. Aviles-Rivero, and Pavlos G. Lagoudakis. A fourier neural operator approach for modelling exciton-polariton condensate systems.���������� ����� �������, 8(1), November 2025

    2025

  29. [29]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.���� ����� �������, 3(3):218–229, March 2021

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.���� ����� �������, 3(3):218–229, March 2021

    2021

  30. [30]

    Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew M

    Zongyi Li, Nikola B. Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew M. Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations.����, abs/2010.08895, 2020

    Pith/arXiv arXiv 2010

  31. [31]

    Wavelet neural operator: a neural operator for parametric partial differential equations, 2022

    Tapas Tripura and Souvik Chakraborty. Wavelet neural operator: a neural operator for parametric partial differential equations, 2022

    2022

  32. [32]

    Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew M

    Zongyi Li, Nikola B. Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew M. Stuart, and Anima Anandkumar. Neural operator: Graph kernel network for partial differential equations.����, abs/2003.03485, 2020

    Pith/arXiv arXiv 2003

  33. [33]

    Spectral networks and locally connected networks on graphs, 2014

    Joan Bruna, Wojciech Zaremba, Arthur Szlam, and Yann LeCun. Spectral networks and locally connected networks on graphs, 2014

    2014

  34. [34]

    Convolutional neural networks on graphs with fast localized spectral filtering.����, abs/1606.09375, 2016

    Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering.����, abs/1606.09375, 2016

    Pith/arXiv arXiv 2016

  35. [35]

    Fourier neural operator with learned deformations for pdes on general geometries.������� �� ������� �������� ��������, 24(388):1–26, 2023

    Zongyi Li, Daniel Zhengyu Huang, Burigede Liu, and Anima Anandkumar. Fourier neural operator with learned deformations for pdes on general geometries.������� �� ������� �������� ��������, 24(388):1–26, 2023

    2023

  36. [36]

    Subhankar Sarkar and Souvik Chakraborty. Spatio-spectral graph neural operator for solving computational mechanics problems on irregular domain and unstructured grid.�������� ������� �� ������� ��������� ��� �����������, 435:117659, 2025

    2025

  37. [37]

    Neuroscience inspired scientific machine learning (part-2): Variable spiking wavelet neural operator, 2023

    Shailesh Garg and Souvik Chakraborty. Neuroscience inspired scientific machine learning (part-2): Variable spiking wavelet neural operator, 2023

    2023

  38. [38]

    Neuroscience inspired scientific machine learning (part-1): Variable spiking neuron for regression, 2023

    Shailesh Garg and Souvik Chakraborty. Neuroscience inspired scientific machine learning (part-1): Variable spiking neuron for regression, 2023

    2023

  39. [39]

    A scalable framework for learning the geometry-dependent solution operators of partial differential equations.������ ������������� �������, 4(12):928–940, December 2024

    Minglang Yin, Nicolas Charon, Ryan Brody, Lu Lu, Natalia Trayanova, and Mauro Maggioni. A scalable framework for learning the geometry-dependent solution operators of partial differential equations.������ ������������� �������, 4(12):928–940, December 2024

    2024

  40. [40]

    Hasan Nasrullah, Istiak Ahmad, Kazuma Kobayashi, and Syed Bahauddin Alam

    Farid Ahmed, Md. Hasan Nasrullah, Istiak Ahmad, Kazuma Kobayashi, and Syed Bahauddin Alam. Enhancing thermo-hydraulic performance in dimpled channels with wavy tape inserts for heat pipe & heat exchanger design with complex energy systems.���� ������� �� ������� �����������, 60:104583, 2024

    2024

  41. [41]

    Optimizing resource allocation for energy efficiency in fog cloud computing environments

    Mandeep Kaur, Rajni Aron, and Shriya Seth. Optimizing resource allocation for energy efficiency in fog cloud computing environments. In���� ���� ���� ������������� ���������� �� ������������� ������� ��� ������� ������������ ������, pages 538–542, 2024

    2024

  42. [42]

    Alsharif, Abu Jahid, Raju Kannadasan, Manish Kumar Singla, Jyoti Gupta, Kottakkaran Sooppy Nisar, Abdel-Haleem Abdel-Aty, and Mun-Kyeom Kim

    Mohammed H. Alsharif, Abu Jahid, Raju Kannadasan, Manish Kumar Singla, Jyoti Gupta, Kottakkaran Sooppy Nisar, Abdel-Haleem Abdel-Aty, and Mun-Kyeom Kim. Survey of energy-efficient fog computing: Techniques and recent advances.������ �������, 13:1739–1763, 2025

    2025

  43. [43]

    Object detection on low-compute edge SoCs: a reproducible benchmark and deployment guidelines.���� ����, 16(1):5875, January 2026

    Chang Kong, Feng Li, Xiaohu Yan, Jinfeng Yang, Peng Mo, Qiuming Luo, and Rui Mao. Object detection on low-compute edge SoCs: a reproducible benchmark and deployment guidelines.���� ����, 16(1):5875, January 2026

    2026

  44. [44]

    Efficient GPT-4V level multimodal large language model for deployment on edge devices.���� �������, 16(1):5509, July 2025

    Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Chi Chen, Haoyu Li, Weilin Zhao, Zhihui He, Qianyu Chen, Ronghua Zhou, Zhensheng Zou, Haoye Zhang, Shengding Hu, Zhi Zheng, Jie Zhou, Jie Cai, Xu Han, Guoyang Zeng, Dahai Li, Zhiyuan Liu, and Maosong Sun. Efficient GPT-4V level multimodal large language model for deployment o...

    2025

  45. [45]

    Cox, Christina Lioma, and Hao Hu

    Xiansong Meng, Deming Kong, Kwangwoong Kim, Qiuchi Li, Po Dong, Ingemar J. Cox, Christina Lioma, and Hao Hu. Digital-analog hybrid matrix multiplication processor for optical neural networks.������ ���������� �����, 16(1), August 2025

    2025

  46. [46]

    Boundary observer design for cascaded ode — hyperbolic pde systems: A matrix inequalities approach.����������, 119:109027, 2020

    Francesco Ferrante, Andrea Cristofaro, and Christophe Prieur. Boundary observer design for cascaded ode — hyperbolic pde systems: A matrix inequalities approach.����������, 119:109027, 2020

    2020

  47. [47]

    Adaptive boundary observers for hyperbolic pdes with application to traffic flow estimation.���� ������������ �� ��������� �������, 69(1):651–658, 2024

    Jiahao Wu, Jingyuan Zhan, and Liguo Zhang. Adaptive boundary observers for hyperbolic pdes with application to traffic flow estimation.���� ������������ �� ��������� �������, 69(1):651–658, 2024

    2024

  48. [48]

    Pde traffic observer validated on freeway data

    Huan Yu, Qijian Gan, Alexandre Bayen, and Miroslav Krstic. Pde traffic observer validated on freeway data. ���� ������������ �� ������� ������� ����������, 29(3):1048–1060, 2021

    2021

  49. [49]

    Continuous monitoring of large-scale generative AI via deterministic knowledge graph structures.����������� �� ��� ���� ��������� ������, 7(1):169–176, November 2025

    Kishor Datta Gupta, Mohd Ariful Haque, Hasmot Ali, Marufa Kamal, Syed Bahauddin Alam, and Moham- mad Ashiqur Rahman. Continuous monitoring of large-scale generative AI via deterministic knowledge graph structures.����������� �� ��� ���� ��������� ������, 7(1):169–176, November 2025

    2025

  50. [50]

    Learning integral operators via neural integral equations.������ ������� ������������, 6(9):1046–1062, August 2024

    Emanuele Zappala, Antonio Henrique de Oliveira Fonseca, Josue Ortega Caro, Andrew Henry Moberly, Michael James Higley, Jessica Cardin, and David van Dijk. Learning integral operators via neural integral equations.������ ������� ������������, 6(9):1046–1062, August 2024

    2024

  51. [51]

    Scabini and Odemir M

    Leonardo F.S. Scabini and Odemir M. Bruno. Structure and performance of fully connected neural networks: Emerging complex network properties.������� �� ����������� ��������� ��� ��� ������������, 615:128585, 2023

    2023

  52. [52]

    Seidman, Georgios Kissas, Paris Perdikaris, and George J

    Jacob H. Seidman, Georgios Kissas, Paris Perdikaris, and George J. Pappas. Nomad: Nonlinear manifold decoders for operator learning, 2022

    2022

  53. [53]

    Regular graph construction for semi-supervised learning.������� �� �������� ���������� ������, 490(1):012022, mar 2014

    Didier A Vega-Oliveros, Lilian Berton, Andre Mantini Eberle, Alneu de Andrade Lopes, and Liang Zhao. Regular graph construction for semi-supervised learning.������� �� �������� ���������� ������, 490(1):012022, mar 2014

    2014

  54. [54]

    Hub-hub connections matter: Improving edge dropout to relieve over-smoothing in graph neural networks.��������������� �������, 270:110556, 2023

    Rui Huang and Ping Li. Hub-hub connections matter: Improving edge dropout to relieve over-smoothing in graph neural networks.��������������� �������, 270:110556, 2023

    2023

  55. [55]

    Pyg documentation, 2026

    PyG Team. Pyg documentation, 2026

    2026

  56. [56]

    AI-driven uncertainty quantification & multi-physics approach to evaluate cladding materials in a microreactor.�������� �� ������� ������, 186:105793, 2025

    Alexander Foutch, Kazuma Kobayashi, Ayodeji Alajo, Dinesh Kumar, and Syed Bahauddin Alam. AI-driven uncertainty quantification & multi-physics approach to evaluate cladding materials in a microreactor.�������� �� ������� ������, 186:105793, 2025

    2025

  57. [57]

    Influence of nuclear data parameters on integral experiment assimilation using cook’s distance

    Dinesh Kumar, Syed Bahauddin Alam, Henrik Sjöstrand, JM Palau, and C De Saint Jean. Influence of nuclear data parameters on integral experiment assimilation using cook’s distance. In��� ��� �� �����������, volume 211, page 07001. EDP Sciences, 2019

    2019

  58. [58]

    Multi-criteria decision making under uncertainties in composite materials selection and design

    Dinesh Kumar, Mariapia Marchi, Syed Bahauddin Alam, Carlos Kavka, Yao Koutsawa, Gaston Rauchs, and Salim Belouettar. Multi-criteria decision making under uncertainties in composite materials selection and design. ��������� ����������, 279:114680, 2022

    2022

  59. [59]

    Hardware based realtime, fast and highly secured communication using FPGA

    Hussain Mohammed Dipu Kabir and Syed Bahauddin Alam. Hardware based realtime, fast and highly secured communication using FPGA. In���� ���� ������������� ���������� �� ����������� ������ ��� ����������� ��������, pages 452–457. IEEE, 2010

    2010

  60. [60]

    Isme Azam, Mohammad Arafat Hussain, A.B.M

    Hussain Mohammed Dipu Kabir, Syed Bahauddin Alam, Md. Isme Azam, Mohammad Arafat Hussain, A.B.M. Rafi Sazzad, Md. Nazmus Sakib, and Md. Abdul Matin. Non-linear down-sampling and signal reconstruc- tion, without folding. In���� ������ ����� �������� ��������� �� �������� �������� ��� ����������, pages 142–146, 2010

    2010

  61. [61]

    Hussain Mohammed Dipu Kabir, Saeed Anwar, Syed Bahauddin Alam, K. M. Sabidur Rahman, Md. Abdul Matin, and Imranul Kabir Chowdhury. Watermarking with fast and highly secured encryption for real-time speech signals. In���� ���� ������������� ���������� �� ����������� ������ ��� ����������� ��������, pages 446–451, 2010

    2010

  62. [62]

    Abdul Matin, and Imranul Kabir Chowdhury

    Hussain Mohammed Dipu Kabir, Syed Bahauddin Alam, Md. Abdul Matin, and Imranul Kabir Chowdhury. A loss-less compression technique for high quality speech signals and its implementation with mpeg-4 als for better compression. In���� ���� ������������� ���������� �� ����������� ������ ��� ����������� ��������, pages 781–785, 2010

    2010

  63. [63]

    Ansys fluent user’s guide, release 2024 r1, 2024

    ANSYS Inc. Ansys fluent user’s guide, release 2024 r1, 2024

    2024

  64. [64]

    Andrew V . Knyazev. Toward the optimal preconditioned eigensolver: Locally optimal block preconditioned conjugate gradient method.���� ������� �� ��������� ���������, 23(2):517–541, 2001

    2001

  65. [65]

    Delta user documentation, 2025

    National Center for Supercomputing Applications (NCSA). Delta user documentation, 2025. Accessed: 2025-11- 10. 33 Computational Sensing of Inaccessible Physical Fields

    2025

  66. [66]

    Staudemeyer and Eric Rothstein Morris

    Ralf C. Staudemeyer and Eric Rothstein Morris. Understanding lstm – a tutorial into long short-term memory recurrent neural networks, 2019

    2019

  67. [67]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2017. 6 Supplementary Information 6.1 Additional Results for Benchmarks We present additional percentile statistics comparing VIRSO with other neural operator models on our chosen bench- marks. We find that on average VIRSO’s distribution has a lower center and similar or narrowe...

    2017