pith. sign in

arxiv: 2202.02710 · v2 · submitted 2022-02-06 · 💻 cs.LG · cs.NA· math.AP· math.NA

Spectrally Adapted Physics-Informed Neural Networks for Solving Unbounded Domain Problems

Mingtao Xia , Lucas B\"ottcher , Tom Chou This is my paper

Pith reviewed 2026-05-24 12:37 UTC · model grok-4.3

classification 💻 cs.LG cs.NAmath.APmath.NA
keywords physics-informed neural networksunbounded domain problemsadaptive spectral methodspartial differential equationsnumerical solversparameter estimationnoisy observations
0
0 comments X

The pith

Spectrally adapted PINNs solve PDEs on unbounded domains by integrating adaptive spectral methods into neural network solvers.

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

The paper seeks to establish that combining adaptive spectral methods with physics-informed neural networks produces solvers capable of handling PDEs where at least one variable ranges over unbounded domains. Standard PINNs struggle with the wide range of scales involved, but the adapted version uses spectral adaptation to resolve dependencies across many orders of magnitude while retaining the network's ease of implementing high-order schemes and extrapolating solutions. This matters for physical applications that naturally involve infinite or semi-infinite domains, such as wave propagation or diffusion at large distances. Examples in the work show the method also supports parameter estimation from noisy data where conventional approaches do not scale well.

Core claim

The central claim is that recently introduced adaptive techniques for spectral methods can be integrated into PINN-based PDE solvers to obtain accurate numerical solutions for unbounded domain problems that cannot be efficiently approximated by standard PINNs, taking advantage of the network's ability to implement high-order schemes and extrapolate at any point.

What carries the argument

The spectrally adapted PINN, which embeds adaptive spectral techniques into the PINN framework to manage the dependence on the unbounded variable.

If this is right

  • PDEs involving variables over unbounded ranges become solvable with resolution across multiple orders of magnitude.
  • High-order numerical schemes for such PDEs can be implemented directly through the neural network structure.
  • Solutions and derivatives can be evaluated at arbitrary points in space and time without additional mesh refinement.
  • Model parameters can be recovered from noisy observations even when the underlying domain is unbounded.

Where Pith is reading between the lines

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

  • The same adaptation principle could be tested on other neural solvers beyond PINNs for problems with infinite boundaries.
  • Scaling to higher-dimensional unbounded domains might become feasible if the spectral adaptation generalizes without added computational cost.
  • Real-time applications such as long-range signal propagation could benefit if the method maintains stability under time evolution.

Load-bearing premise

Adaptive spectral techniques can be added to PINNs without harming the network's extrapolation ability or high-order scheme handling.

What would settle it

A direct comparison on a known analytic unbounded-domain PDE where the spectrally adapted PINN shows no accuracy gain or convergence improvement over standard PINNs across increasing domain sizes.

Figures

Figures reproduced from arXiv: 2202.02710 by Lucas B\"ottcher, Mingtao Xia, Tom Chou.

Figure 1
Figure 1. Figure 1: Solving unbounded domain problems with spectrally adapted physics￾informed neural networks for functions uN(x, t) that can be expressed as a spectral expansion uN(x, t) = PN i=0 wi(t)φi(x). (a) An example of a function uN(x, t) plotted at three different time points. (b) Decaying behavior of a corresponding basis function element φi(x). (c) PDEs in unbounded domains can be solved by combining spectral deco… view at source ↗
Figure 2
Figure 2. Figure 2: Example 1: Function approximation. Approximation of the target function Eq. 3 using both standard neural networks and a spectral multi-output neural network that learns the coefficients wi(t; Θ) in the spectral expansion Eq. 1. Comparison of the approximation error using a spectral multi-output neural network (red) with the error incurred when using a standard neural-network function approximator (black). … view at source ↗
Figure 3
Figure 3. Figure 3: Example 2: Solving Eq. 11 in a bounded domain. L 2 errors, frequency indicators, and expansion order associated with the numerical solution of Eq. 11 using the adaptive s-PINN method with a timestep ∆t = 0.01. (a) In a bounded domain, the s￾PINNs, with and without the adaptive spectral technique, have smaller errors than the standard PINN (black). Moreover, the s-PINN method combined with a p-adaptive tech… view at source ↗
Figure 4
Figure 4. Figure 4: Example 3: Solving Eq. 13 in an unbounded domain. L 2 error, frequency indicator, and expansion order associated with the numerical solution of Eq. 13 using the s-PINN method combined with the spectral scaling technique. (a) The s-PINN method with the scaling technique (red) has a smaller error than the s-PINN without scaling (blue). The higher accuracy of the adaptive s-PINN is a consequence of maintainin… view at source ↗
Figure 5
Figure 5. Figure 5: Example 4: Solving a higher dimensional unbounded domain PDE (Eq. 16). L 2 error, scaling factor, and frequency indicators associated with the numerical solution of Eq. 16 using s-PINNs, with and without dynamic scaling. (a) L 2 error as a function of time. The s-PINNs that are equipped with the scaling technique (red) achieve higher accuracy than those without (black). (b) The scaling factors βx (blue) an… view at source ↗
Figure 6
Figure 6. Figure 6: Example 6: Solving the Schr ¨odinger equation (Eq. 22) in an unbounded domain. Approximation error, scaling factor, displacement, and expansion order associated with the numerical solution of Eq. 22 using adaptive (red) and non-adaptive (black) s-PINNs. (a) Errors for numerically solving Eq. 22 with and without adaptive techniques. (b) The change of the scaling factor which decreases over time as the solut… view at source ↗
Figure 7
Figure 7. Figure 7: Example 8: Parameter (diffusivity) inference. The parameter κ inferred within successive time windows of ∆t = 0.1, the SSE error Eq. 29, the scaling factor, and the frequency indicators associated with solving Eq. 31, for different noise levels σ 2 . Here, the SSE was minimized to find the estimate θˆ ≡ κˆ and the solutions uN at intermediate timesteps tj + cs∆t. (a, b) Smaller σ 2 leads to smaller SSE Eq.… view at source ↗
Figure 8
Figure 8. Figure 8: Example 9: Source recovery. SSE0 plotted against the reconstructed heat source khNk2 as given by 35, as a function of λ for various values of σ 2 (an “L￾curve”). When λ is large, the norm of the reconstructed heat source khNk2 always tends to decrease while the “error” SSE0 tends to increase. When λ = 10−1 , khNk2 is small and the SSE0 is large. A moderate λ ∈ [10−2 , 10−3 ] could reduce the error SSE0, co… view at source ↗
read the original abstract

Solving analytically intractable partial differential equations (PDEs) that involve at least one variable defined on an unbounded domain arises in numerous physical applications. Accurately solving unbounded domain PDEs requires efficient numerical methods that can resolve the dependence of the PDE on the unbounded variable over at least several orders of magnitude. We propose a solution to such problems by combining two classes of numerical methods: (i) adaptive spectral methods and (ii) physics-informed neural networks (PINNs). The numerical approach that we develop takes advantage of the ability of physics-informed neural networks to easily implement high-order numerical schemes to efficiently solve PDEs and extrapolate numerical solutions at any point in space and time. We then show how recently introduced adaptive techniques for spectral methods can be integrated into PINN-based PDE solvers to obtain numerical solutions of unbounded domain problems that cannot be efficiently approximated by standard PINNs. Through a number of examples, we demonstrate the advantages of the proposed spectrally adapted PINNs in solving PDEs and estimating model parameters from noisy observations in unbounded domains.

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

0 major / 3 minor

Summary. The manuscript proposes spectrally adapted PINNs that integrate adaptive spectral techniques (e.g., mapped bases) into the PINN residual loss to solve PDEs on unbounded domains. It demonstrates advantages over standard PINNs via numerical examples for both forward PDE solves and inverse parameter estimation from noisy data.

Significance. If the results hold, the approach extends PINNs to a class of problems that are common in applications but difficult for standard formulations due to the need to resolve behavior over many orders of magnitude. The concrete constructions (mapped bases inside the loss) and consistent numerical demonstrations in the examples constitute a practical contribution; the method preserves the extrapolation and high-order capabilities of PINNs while adding spectral adaptivity.

minor comments (3)
  1. [Section 3] The integration step (how the adaptive spectral basis is inserted into the PINN loss) is described at a high level; adding an explicit equation or pseudocode block would improve reproducibility.
  2. [Section 4] Several example figures compare solution profiles but lack quantitative error tables versus standard PINNs or other unbounded-domain methods; adding such metrics would strengthen the advantage claims.
  3. [Abstract / Introduction] The abstract and introduction refer to 'recently introduced adaptive techniques' without a specific citation in the opening paragraphs; adding the reference at first mention would aid readers.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the recommendation of minor revision. The recognition that the concrete constructions and numerical demonstrations constitute a practical contribution is appreciated. No specific major comments appear in the report.

Circularity Check

0 steps flagged

No significant circularity; method combines independent existing techniques

full rationale

The paper presents a hybrid numerical method that integrates adaptive spectral techniques (cited as recently introduced) into the PINN framework for unbounded-domain PDEs. The central construction is described as a concrete combination: mapped bases inside the residual loss, with advantages shown through explicit numerical examples for forward and inverse problems. No derivation step reduces by construction to a fitted parameter renamed as prediction, no self-definitional loop appears in the equations, and no load-bearing uniqueness theorem or ansatz is imported solely via self-citation. The reported demonstrations are external to any internal fit, satisfying the criteria for a self-contained, non-circular contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the integration itself is the contribution.

pith-pipeline@v0.9.0 · 5717 in / 986 out tokens · 14573 ms · 2026-05-24T12:37:23.256439+00:00 · methodology

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

65 extracted references · 65 canonical work pages · 1 internal anchor

  1. [1]

    In a dataset ( xs, ts, us), s ∈ {1,..., n}, xs are values of the sampled “spatial” variable x which can be defined in an unbounded domain

    Combining Spectral Methods with Neural Networks In this section, we first introduce the basic features of func tion approximators that rely on neural networks and spectral methods designed to handle v ariables that are defined in unbounded domains. In a dataset ( xs, ts, us), s ∈ {1,..., n}, xs are values of the sampled “spatial” variable x which can be defi...

    work page

  2. [2]

    We apply s-PINNs to numerically solve PDEs, and in pa rticular, spatiotemporal PDEs in unbounded domains for which standard PINN approache s cannot be directly applied

    Application to Solving PDEs In this section, we show that spectrally adapted neural netw orks can be combined with physics-informed neural networks (PINNs) which we shall ca ll spectrally adapted PINNs (s- PINNs). We apply s-PINNs to numerically solve PDEs, and in pa rticular, spatiotemporal PDEs in unbounded domains for which standard PINN approache s ca...

    work page 2098

  3. [3]

    25 admits the analytical solution u(x, t) = sin x√ t + 1 exp [ − x2 4(t + 1) ]

    sin x] (t + 1)−3/ 2 exp[− x2 4(t+1) ], Eq. 25 admits the analytical solution u(x, t) = sin x√ t + 1 exp [ − x2 4(t + 1) ] . (26) We solve Eq. 25 in the weak form by multiplying any test functi on v ∈H1(R) on both sides and integrating by parts to obtain (∂tu, v) = −(∂xu,∂xv) + ( f, v), ∀v ∈H1(R). (27) Spectrally Adapted Neural Networks 18 T able 3. Exampl...

    work page

  4. [4]

    L- curve

    Parameter Inference and Source Reconstruction As with standard PINN approaches, s-PINNs can also be used fo r parameter inference in PDE models or reconstructing unknown sources in a physical mode l. Assuming observational data at uniform time intervals t j = j∆t associated with a partially known underlying PDE model, Spectrally Adapted Neural Networks 20...

    work page

  5. [5]

    The underlying f eature that we exploit is the physical di fferences across classes of data

    Summary and Conclusions In this paper, we propose an approach that blends standard PI NN algorithms with adaptive spectral methods and show through examples that this hybrid approach can be applied to a wide variety of data-driven problems including function a pproximation, solving PDEs, parameter inference, and model selection. The underlying f eature th...

    work page

  6. [6]

    +” and “ –

    can be incorporated when applying our s-PINN for inverse-type PDE discovery problems. Finally, one can incorporate a recently proposed Bayesian- PINN (B-PINN) [59] method into our s-PINN method to quantify uncertainty when solving inve rse problems under noisy data. Spectrally Adapted Neural Networks 26 T able 6. Advantages and disadvantages of traditiona...

    work page

  7. [7]

    Approximation capabilities of multilayer feedforward networks

    Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural Networks, 4(2):251– 257, 1991

    work page 1991

  8. [8]

    Mini mum width for universal approximation

    Sejun Park, Chulhee Y un, Jaeho Lee, and Jinwoo Shin. Mini mum width for universal approximation. In International Conference on Learning Representations , 2020

    work page 2020

  9. [9]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problem s involving nonlinear partial di fferential equations

    Maziar Raissi, Paris Perdikaris, and George E Karniadak is. Physics-informed neural networks: A deep learning framework for solving forward and inverse problem s involving nonlinear partial di fferential equations. Journal of Computational Physics , 378:686–707, 2019

    work page 2019

  10. [10]

    Physics-informed machine learning

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Pari s Perdikaris, Sifan Wang, and Liu Y ang. Physics-informed machine learning. Nature Reviews Physics, 3(6):422–440, 2021

    work page 2021

  11. [11]

    Neural ordinary di fferential equation control of dynamics on graphs

    Thomas Asikis, Lucas B¨ ottcher, and Nino Antulov-Fantulin. Neural ordinary di fferential equation control of dynamics on graphs. Physical Review Research (in press), 2022

    work page 2022

  12. [12]

    AI Pontryagin or how neural networks learn to control dynamical systems

    Lucas B¨ ottcher, Nino Antulov-Fantulin, and Thomas Asikis. AI Pontryagin or how neural networks learn to control dynamical systems. Nature Communications, 2021

    work page 2021

  13. [13]

    Near-optimal contro l of dynamical systems with neural ordinary differential equations

    Lucas B¨ ottcher and Thomas Asikis. Near-optimal contro l of dynamical systems with neural ordinary differential equations. Machine Learning: Science and T echnology, 3(4):045004, 2022

    work page 2022

  14. [14]

    Neural network control of robot manipulators and non-linear systems

    FW Lewis, Suresh Jagannathan, and A ydin Y esildirak. Neural network control of robot manipulators and non-linear systems. CRC Press, 2020

    work page 2020

  15. [15]

    Regularization for Deep Learning: A Taxonomy

    Jan Kukaˇ cka, Vladimir Golkov, and Daniel Cremers. Regularization for deep learning: A taxonomy. arXiv preprint arXiv:1710.10686, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  16. [16]

    Deep Lagrangian netw orks: Using physics as model prior for deep learning

    M Lutter, C Ritter, and Jan Peters. Deep Lagrangian netw orks: Using physics as model prior for deep learning. In International Conference on Learning Representations . OpenReview.net, 2019

    work page 2019

  17. [17]

    Modeling system dynamics with physics-informed neural net works based on Lagrangian mechanics

    Manuel A Roehrl, Thomas A Runkler, V eronika Brandtstet ter, Michel Tokic, and Stefan Obermayer. Modeling system dynamics with physics-informed neural net works based on Lagrangian mechanics. IF AC-PapersOnLine, 53(2):9195–9200, 2020

    work page 2020

  18. [18]

    Symplectic ODE-net: Learning Hamiltonian dynamics with control

    Y aofeng Desmond Zhong, Biswadip Dey, and Amit Chakrabo rty. Symplectic ODE-net: Learning Hamiltonian dynamics with control. In International Conference on Learning Representations , 2019

    work page 2019

  19. [19]

    V ariational physics-informed neural networks for solving partial di fferential equations

    Ehsan Kharazmi, Zhongqiang Zhang, and George Em Karnia dakis. V ariational physics-informed neural networks for solving partial di fferential equations. arXiv preprint arXiv:1912.00873 , 2019

    work page arXiv 1912

  20. [20]

    Extended phys ics-informed neural networks (xpinns): A generalized space-time domain decomposition based deep l earning framework for nonlinear partial differential equations

    Ameya D Jagtap and George Em Karniadakis. Extended phys ics-informed neural networks (xpinns): A generalized space-time domain decomposition based deep l earning framework for nonlinear partial differential equations. Communications in Computational Physics , 28(5):2002–2041, 2020

    work page 2002

  21. [21]

    Physics-informed neural operator for learning partial differential equations, 2024

    Zongyi Li, Hongkai Zheng, Nikola Kovachki, David Jin, H aoxuan Chen, Burigede Liu, Kamyar Azizzadenesheli, and Anima Anandkumar. Physics-informed neural operator for learning partial differential equations. arXiv preprint arXiv:2111.03794 , 2021

    work page arXiv 2021

  22. [22]

    Physics-informed neural networks for high- speed flows

    Zhiping Mao, Ameya D Jagtap, and George Em Karniadakis. Physics-informed neural networks for high- speed flows. Computer Methods in Applied Mechanics and Engineering , 360:112789, 2020

    work page 2020

  23. [23]

    A physics-informed neural network framework for PDEs on 3D surfaces: Time independent problems

    Zhiwei Fang and Justin Zhan. A physics-informed neural network framework for PDEs on 3D surfaces: Time independent problems. IEEE Access, 8:26328–26335, 2019

    work page 2019

  24. [24]

    Physics-informed neural networks for power systems

    George S Misyris, Andreas V enzke, and Spyros Chatzivas ileiadis. Physics-informed neural networks for power systems. In 2020 IEEE Power & Energy Society General Meeting (PESGM) , pages 1–5. IEEE, 2020

    work page 2020

  25. [25]

    Physics- informed neural networks for cardiac activation mapping

    Francisco Sahli Costabal, Yibo Y ang, Paris Perdikaris , Daniel E Hurtado, and Ellen Kuhl. Physics- informed neural networks for cardiac activation mapping. Frontiers in Physics, 8:42, 2020

    work page 2020

  26. [26]

    A generic physic s-informed neural network-based constitutive model for soft biological tissues

    Minliang Liu, Liang Liang, and Wei Sun. A generic physic s-informed neural network-based constitutive model for soft biological tissues. Computer methods in applied mechanics and engineering, 372:113402, 2020. Spectrally Adapted Neural Networks 28

    work page 2020

  27. [27]

    Computational Statistical Physics

    Lucas B¨ ottcher and Hans J Herrmann. Computational Statistical Physics . Cambridge University Press, 2021

    work page 2021

  28. [28]

    Modeling deformed t ransmission lines for continuous strain sensing applications

    Stefan H Strub and Lucas B¨ ottcher. Modeling deformed t ransmission lines for continuous strain sensing applications. Measurement Science and T echnology, 31(3):035109, 2019

    work page 2019

  29. [29]

    Algebraic damping in the one-dimensional Vlasov equation

    Julien Barr´ e, Alain Olivetti, and Y oshiyuki Y Y amaguc hi. Algebraic damping in the one-dimensional Vlasov equation. Journal of Physics A: Mathematical and Theoretical , 44(40):405502, 2011

    work page 2011

  30. [30]

    Stability and error analysis for a second-order fast approximation of the one-dimensional Schr¨ odinger equati on under absorbing boundary conditions

    Buyang Li, Jiwei Zhang, and Chunxiong Zheng. Stability and error analysis for a second-order fast approximation of the one-dimensional Schr¨ odinger equati on under absorbing boundary conditions. SIAM Journal on Scientific Computing , 40(6):A4083–A4104, 2018

    work page 2018

  31. [31]

    PDE models o f adder mechanisms in cellular proliferation

    Mingtao Xia, Chris D Greenman, and Tom Chou. PDE models o f adder mechanisms in cellular proliferation. SIAM Journal on Applied Mathematics , 80(3):1307–1335, 2020

    work page 2020

  32. [32]

    Kinetic theory for structured populations: application to stochastic sizer-timer models of cell proliferation

    Mingtao Xia and Tom Chou. Kinetic theory for structured populations: application to stochastic sizer-timer models of cell proliferation. Journal of Physics A: Mathematical and Theoretical , 2021

    work page 2021

  33. [33]

    Real-space observation of emergent magnet ic monopoles and associated Dirac strings in artificial Kagom´ e spin ice

    Elena Mengotti, Laura J Heyderman, Arantxa Fraile Rodr ´ ıguez, Frithjof Nolting, Remo V H¨ ugli, and Hans- Benjamin Braun. Real-space observation of emergent magnet ic monopoles and associated Dirac strings in artificial Kagom´ e spin ice. Nature Physics, 7(1):68–74, 2011

    work page 2011

  34. [34]

    Artificial Kagom´ e spin ice: dimensional reductio n, avalanche control and emergent magnetic monopoles

    RV H¨ ugli, G Du ff, B O’Conchuir, E Mengotti, A Fraile Rodr´ ıguez, F Nolting, L J Heyderman, and HB Braun. Artificial Kagom´ e spin ice: dimensional reductio n, avalanche control and emergent magnetic monopoles. Philosophical Transactions of the Royal Society A: Mathema tical, Physical and Engineering Sciences, 370(1981):5767–5782, 2012

    work page 1981

  35. [35]

    E fficient scaling and moving techniques for spectral methods in unbounded domains

    Mingtao Xia, Sihong Shao, and Tom Chou. E fficient scaling and moving techniques for spectral methods in unbounded domains. SIAM Journal on Scientific Computing , 43(5):A3244–A3268, 2021

    work page 2021

  36. [36]

    A frequency-dep endent p-adaptive technique for spectral methods

    Mingtao Xia, Sihong Shao, and Tom Chou. A frequency-dep endent p-adaptive technique for spectral methods. Journal of Computational Physics , 446:110627, 2021

    work page 2021

  37. [37]

    Spectral methods: algorithms, analysis and applications , volume 41

    Jie Shen, Tao Tang, and Li-Lian Wang. Spectral methods: algorithms, analysis and applications , volume 41. Springer Science & Business Media, 2011

    work page 2011

  38. [38]

    Spectral methods in MATLAB

    Lloyd N Trefethen. Spectral methods in MATLAB . SIAM, 2000

    work page 2000

  39. [39]

    Taylor expansion of the accumulated rounding error

    Seppo Linnainmaa. Taylor expansion of the accumulated rounding error. BIT Numerical Mathematics , 16(2):146–160, 1976

    work page 1976

  40. [40]

    Automati c differentiation in PyTorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chan an, Edward Y ang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automati c differentiation in PyTorch. 2017

    work page 2017

  41. [41]

    M ultilayer feedforward networks are universal approximators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. M ultilayer feedforward networks are universal approximators. Neural Networks, 2(5):359–366, 1989

    work page 1989

  42. [42]

    Batch normalization: Acceleratin g deep network training by reducing internal covariate shift

    Sergey Io ffe and Christian Szegedy. Batch normalization: Acceleratin g deep network training by reducing internal covariate shift. In International Conference on Machine Learning , pages 448–456. PMLR, 2015

    work page 2015

  43. [43]

    Rati onal spectral methods for PDEs involving fractional Laplacian in unbounded domains

    Tao Tang, Li-Lian Wang, Huifang Y uan, and Tao Zhou. Rati onal spectral methods for PDEs involving fractional Laplacian in unbounded domains. SIAM Journal on Scientific Computing , 42(2):A585–A611, 2020

    work page 2020

  44. [44]

    Automatic di fferentiation in machine learning: a survey

    Atilim Gunes Baydin, Barak A Pearlmutter, Alexey Andre yevich Radul, and Je ffrey Mark Siskind. Automatic di fferentiation in machine learning: a survey. Journal of Machine Learning Research , 18, 2018

    work page 2018

  45. [45]

    Message passing neural PDE solvers

    Johannes Brandstetter, Daniel E Worrall, and Max Welli ng. Message passing neural PDE solvers. In International Conference on Learning Representations , 2021

    work page 2021

  46. [46]

    Fourier neural operator for parame tric partial di fferential equations

    Zongyi Li, Nikola Borislavov Kovachki, Kamyar Azizzad enesheli, Kaushik Bhattacharya, Andrew Stuart, Anima Anandkumar, et al. Fourier neural operator for parame tric partial di fferential equations. In International Conference on Learning Representations , 2020

    work page 2020

  47. [47]

    Sparse spectral approximati ons of high-dimensional problems based on hyperbolic cross

    Jie Shen and Li-Lian Wang. Sparse spectral approximati ons of high-dimensional problems based on hyperbolic cross. SIAM Journal on Numerical Analysis , 48(3):1087–1109, 2010

    work page 2010

  48. [48]

    On the optim ization of deep networks: Implicit acceleration by overparameterization

    Sanjeev Arora, Nadav Cohen, and Elad Hazan. On the optim ization of deep networks: Implicit acceleration by overparameterization. In Jennifer Dy and A ndreas Krause, editors, Proceedings of the Spectrally Adapted Neural Networks 29 35th International Conference on Machine Learning , volume 80 of Proceedings of Machine Learning Research, pages 244–253, 2018

    work page 2018

  49. [49]

    How m uch over-parameterization is su fficient to learn deep ReLU networks? In International Conference on Learning Representations , 2020

    Zixiang Chen, Y uan Cao, Difan Zou, and Quanquan Gu. How m uch over-parameterization is su fficient to learn deep ReLU networks? In International Conference on Learning Representations , 2020

    work page 2020

  50. [50]

    Mousa J. Huntul. Identification of the timewise thermal conductivity in a 2D heat equation from local heat flux conditions. Inverse Problems in Science and Engineering , 29(7):903–919, 2021

    work page 2021

  51. [51]

    Inverse problems for the heat-conduction equation with nonlocal boundary conditions

    NI Ivanchov. Inverse problems for the heat-conduction equation with nonlocal boundary conditions. Ukrainian Mathematical Journal, 45(8):1186–1192, 1993

    work page 1993

  52. [52]

    The determination of a coe fficient in a parabolic di fferential equation: Part i

    B Frank Jones Jr. The determination of a coe fficient in a parabolic di fferential equation: Part i. existence and uniqueness. Journal of Mathematics and Mechanics , pages 907–918, 1962

    work page 1962

  53. [53]

    N. Y a. Beznoshchenko. On finding a coe fficient in a parabolic equation. Differential Equations, 10:24–35, 1974

    work page 1974

  54. [54]

    A meshless met hod for solving an inverse spacewise- dependent heat source problem

    Liang Y an, Feng-Lian Y ang, and Chu-Li Fu. A meshless met hod for solving an inverse spacewise- dependent heat source problem. Journal of Computational Physics , 228(1):123–136, 2009

    work page 2009

  55. [55]

    Inverse problem of time-dependent heat sources numerical reconstruction

    Liu Y ang, Mehdi Dehghan, Jian-Ning Y u, and Guan-Wei Luo . Inverse problem of time-dependent heat sources numerical reconstruction. Mathematics and Computers in Simulation , 81(8):1656–1672, 2011

    work page 2011

  56. [56]

    A simplified Tikhonov regulariza tion method for determining the heat source

    Fan Y ang and Chu-Li Fu. A simplified Tikhonov regulariza tion method for determining the heat source. Applied Mathematical Modelling , 34(11):3286–3299, 2010

    work page 2010

  57. [57]

    Toward an artificial intellig ence physicist for unsupervised learning

    Tailin Wu and Max Tegmark. Toward an artificial intellig ence physicist for unsupervised learning. Physical Review E, 100(3):033311, 2019

    work page 2019

  58. [58]

    Determination of an unknown heat so urce from overspecified boundary data

    John Rozier Cannon. Determination of an unknown heat so urce from overspecified boundary data. SIAM Journal on Numerical Analysis , 5(2):275–286, 1968

    work page 1968

  59. [59]

    A variational meth od for identifying a spacewise-dependent heat source

    B Tomas Johansson and Daniel Lesnic. A variational meth od for identifying a spacewise-dependent heat source. IMA Journal of Applied Mathematics , 72(6):748–760, 2007

    work page 2007

  60. [60]

    A unified approach to identifying an unknown spacewise dependent source in a variable coe fficient parabolic equation from final and integral overdeterm inations

    Alemdar Hasanov and Burhan Pekta cc. A unified approach to identifying an unknown spacewise dependent source in a variable coe fficient parabolic equation from final and integral overdeterm inations. Applied Numerical Mathematics , 78:49–67, 2014

    work page 2014

  61. [61]

    PDE- net: Learning PDEs from data

    Zichao Long, Yiping Lu, Xianzhong Ma, and Bin Dong. PDE- net: Learning PDEs from data. In International Conference on Machine Learning , pages 3208–3216. PMLR, 2018

    work page 2018

  62. [62]

    Deep hidden physics models: Deep learni ng of nonlinear partial di fferential equations

    Maziar Raissi. Deep hidden physics models: Deep learni ng of nonlinear partial di fferential equations. Journal of Machine Learning Research , 19(1):932–955, 2018

    work page 2018

  63. [63]

    Recipes for when physics fails: Recovering robust learning of physics informed neural netw orks

    Chandrajit Bajaj, Luke McLennan, Timothy Andeen, and A vik Roy. Recipes for when physics fails: Recovering robust learning of physics informed neural netw orks. Machine Learning: Science and T echnology, 2023

    work page 2023

  64. [64]

    Noise-aware physics- informed machine learning for robust pde discovery

    Pongpisit Thanasutives, Takashi Morita, Masayuki Num ao, and Ken-ichi Fukui. Noise-aware physics- informed machine learning for robust pde discovery. Machine Learning: Science and T echnology , 4:015009, 2022

    work page 2022

  65. [65]

    B-PINN s: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy dat a

    Liu Y ang, Xuhui Meng, and George Em Karniadakis. B-PINN s: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy dat a. Journal of Computational Physics , 425:109913, 2021

    work page 2021