pith. sign in

arxiv: 2606.07579 · v1 · pith:3JUP47C3new · submitted 2026-05-26 · ⚛️ physics.comp-ph · math-ph· math.MP

Exact Boundary Enforcement Along Implicit Geometries for Physics-Informed, Deep Learning Problems in Continuum Mechanics

Cody Rucker , Brittany A. Erickson This is my paper

Pith reviewed 2026-06-29 14:09 UTC · model grok-4.3

classification ⚛️ physics.comp-ph math-phmath.MP
keywords physics-informed neural networksboundary enforcementelastodynamicsplane straincontinuum mechanicsimplicit geometriestraction conditions
0
0 comments X

The pith

PINNs achieve higher accuracy on first-order plane strain elastodynamics, but accuracy and training time trade off with the mix of hard versus soft boundary enforcement.

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

The paper tests how hard and soft enforcement of boundary conditions affects physics-informed neural network solutions to initial-boundary-value problems in continuum mechanics. Using interpolation over implicit representations to hard-enforce traction conditions on arbitrary domains, the work compares first- and second-order formulations of plane-strain elastodynamics. It finds first-order versions deliver higher relative accuracy overall, while the balance of hard and soft boundaries produces a clear tradeoff: more soft enforcement raises accuracy at the cost of longer training runs, and more hard enforcement shortens runs but lowers accuracy. This matters because forward models must be both reliable and efficient when used inside inversion techniques.

Core claim

PINNs achieve a higher relative accuracy when solving the first-order plane strain problem and a tradeoff exists between final relative error and total training runtime, characterized by the number of hard and soft boundaries, where all-soft enforcement yields greater accuracy with longer runtime and all-hard enforcement yields lesser accuracy with shorter runtime.

What carries the argument

Interpolation of boundary data over implicit boundary representations for hard-enforcement of traction conditions on arbitrary domains.

If this is right

  • First-order formulations of the governing equations produce higher relative accuracy than second-order formulations.
  • Raising the fraction of soft boundary enforcement improves final accuracy.
  • Raising the fraction of hard boundary enforcement reduces total training runtime.
  • The observed accuracy-runtime tradeoff is governed by the specific count of hard versus soft boundaries.

Where Pith is reading between the lines

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

  • The tradeoff may guide selection of enforcement strategy when PINNs are embedded in larger optimization loops with limited compute.
  • The same hard/soft counting approach could be tested on time-dependent three-dimensional problems to see whether the pattern generalizes.
  • Pairing the method with adaptive collocation-point sampling might shift the observed accuracy-runtime curve.

Load-bearing premise

Interpolating boundary data over implicit boundary representations accurately captures the geometry and traction conditions for arbitrary domains without significant discretization or approximation error.

What would settle it

Solve the same plane-strain problem on a rectangular domain with a known analytical solution using varying ratios of hard and soft boundaries and check whether relative error indeed decreases and runtime increases as the fraction of soft boundaries rises.

Figures

Figures reproduced from arXiv: 2606.07579 by Brittany A. Erickson, Cody Rucker.

Figure 1
Figure 1. Figure 1: Construction of an approximate distance function for the line segment joining points A = (0, 0) and B = (1, 1). a) depicts the signed distance function (10) to the line AB; b) depicts the trimming function (11); and c) depicts the resulting approximate distance function (12) to line segment AB. 3.3. Approximate distance to a trimmed line segment The line segment joining points x1 = (x1, y1) and x2 = (x2, y… view at source ↗
Figure 2
Figure 2. Figure 2: A domain schematic showing the types of boundary conditions considered at each boundary for first and second-order problems. Conditions at Left boundary are always chosen to be the same type as the conditions at the right boundary (e.g., Left and right both have displacement conditions or both have traction conditions when considering the second-order problem). employ the L 2−norm on a function space of ve… view at source ↗
Figure 3
Figure 3. Figure 3: a) Total loss (MSE) and b) relative L2-error during network training when solving the Second-order plane strain problem with displacement conditions imposed at each spatial boundary. relative error, the case of all-hard enforcement shows superior relative accuracy over fewer training iterations. 5.1.2. Displacements and Tractions Now, we repeat the same simulation as in Section 5.1.1 but this time we speci… view at source ↗
Figure 4
Figure 4. Figure 4: a) Total loss (MSE) and b) relative L2 -error during network training when solv￾ing the second-order plane strain problem with displacement conditions imposed at top and bottom boundaries and traction conditions imposed at left and right boundaries. which may not satisfy the governing equations and constitutive law (39) so we compute appropriate source terms for each and specify boundary data accord￾ingly.… view at source ↗
Figure 5
Figure 5. Figure 5: a) Total loss (MSE) and b) relative L2 -error during network training when solving the first-order plane strain problem with velocity conditions imposed at each spatial boundary. 5.2.1. All Velocities Continuing in a fashion similar to the second-order numerical tests, we will first consider the case where velocity is specified at every spatial boundary so that ΓD = ΓL ∪ ΓB ∪ ΓR ∪ ΓT while the traction bou… view at source ↗
Figure 6
Figure 6. Figure 6: a) Total loss (MSE) and b) relative L2-error during network training when solv￾ing the first-order plane strain problem with velocity conditions imposed at top and bottom boundaries and traction conditions imposed on the stress variable at the left and right bound￾aries. nearly half that of implementing soft-enforcement at every boundary. Our findings highlight an important trade-off between the choice of … view at source ↗
read the original abstract

Solutions to well-posed problems in continuum mechanics are continuously dependent upon prescribed boundary conditions. Because of this, variations in the enforcement of boundary data can impact the reliability of inversion techniques that rely on efficient and accurate forward models. To this end, it is necessary to understand how specific boundary implementation techniques can affect the performance of a given forward model. Our work focuses on the impact that key modeling decisions have on physics-informed neural network (PINN) solutions for initial boundary value problems in continuum mechanics. By interpolating boundary data over implicit boundary representations, we measure the performance of a physics-informed neural network across different configurations of soft and hard boundary enforcement. We target the problem of elastodynamic plane-strain and present a method of hard-enforcement of traction conditions over arbitrary, implicitly-defined, domain boundaries considering both first and second order formulations of the governing equations. We show that PINNs achieve a higher relative accuracy when solving the first-order plane strain problem and we observe a tradeoff between the final relative error and the total run time to complete training. This tradeoff is characterized by the number of hard and soft boundaries where, in the extremes, all soft-enforcement results in greater accuracy with a longer run time, while all hard-enforcement leads to lesser accuracy and a shorter run time.

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

1 major / 2 minor

Summary. The manuscript investigates the impact of soft versus hard boundary enforcement on PINN solutions to elastodynamic plane-strain problems. Using interpolation of boundary data over implicit representations, it compares first- and second-order formulations across mixed hard/soft configurations and reports that the first-order formulation yields higher relative accuracy while the number of hard versus soft boundaries produces an accuracy-runtime tradeoff (all-soft yields highest accuracy but longest training; all-hard yields lowest accuracy but shortest training).

Significance. If the interpolation-based hard enforcement can be shown to introduce negligible geometry/traction error relative to the reported residuals, the empirical comparison would usefully inform boundary-enforcement choices for PINN forward models in continuum mechanics. The work supplies a concrete method for hard-enforcing traction conditions on arbitrary implicit domains and documents a practical accuracy-runtime tradeoff, both of which could guide practitioners even if the quantitative bounds require strengthening.

major comments (1)
  1. [Abstract] Abstract, paragraph on method for hard-enforcement of traction conditions: the central accuracy-runtime tradeoff claim assumes the interpolation procedure enforces boundaries exactly. No a-priori error bound on the interpolation nor a quantitative check (e.g., boundary residual evaluated on a manufactured geometry) is supplied to demonstrate that the interpolation error remains orders of magnitude below the reported relative errors. Without this verification the 'hard' cases are not demonstrably stricter than the soft cases, undermining the tradeoff interpretation.
minor comments (2)
  1. [Abstract] The abstract asserts performance differences and a tradeoff but supplies no error bars, dataset sizes, convergence plots, or explicit description of how relative accuracy was computed.
  2. Consider adding a table or figure that reports the exact number of hard/soft boundaries, final relative errors, and wall-clock times for each configuration so the tradeoff can be inspected quantitatively.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract, paragraph on method for hard-enforcement of traction conditions: the central accuracy-runtime tradeoff claim assumes the interpolation procedure enforces boundaries exactly. No a-priori error bound on the interpolation nor a quantitative check (e.g., boundary residual evaluated on a manufactured geometry) is supplied to demonstrate that the interpolation error remains orders of magnitude below the reported relative errors. Without this verification the 'hard' cases are not demonstrably stricter than the soft cases, undermining the tradeoff interpretation.

    Authors: We agree that the current manuscript lacks an explicit verification that the interpolation error is negligible relative to the reported residuals. In the revision we will add a quantitative check (boundary residual evaluated on a manufactured geometry) demonstrating that the interpolation error lies orders of magnitude below the relative errors shown in the results. This addition will confirm that the hard-enforcement cases are indeed stricter and will support the accuracy-runtime tradeoff interpretation. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical measurement study with independent experimental results.

full rationale

The paper reports direct measurements of PINN training accuracy and runtime across hard/soft boundary configurations for elastodynamic plane-strain problems. No equations, derivations, or 'predictions' are presented that reduce reported quantities to fitted parameters by construction, nor does any load-bearing claim rest on self-citation chains. The central tradeoff observation follows from the experimental design itself and is not forced by redefinition or imported uniqueness results. The work is self-contained against external benchmarks as an empirical study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no equations, no explicit free parameters, and no invented entities; ledger therefore empty.

pith-pipeline@v0.9.1-grok · 5762 in / 1133 out tokens · 28146 ms · 2026-06-29T14:09:07.052343+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

72 extracted references · 53 canonical work pages

  1. [1]

    Variationally correct neural residual regression for parametric pdes: On the viability of controlled ac- curacy

    Bachmayr, M., Dahmen, W., Oster, M., 2025. Variationally correct neural residual regression for parametric pdes: On the viability of controlled ac- curacy. IMA Journal of Numerical Analysis , draf073doi:10.1093/imanum/ draf073

    work page doi:10.1093/imanum/ 2025

  2. [2]

    Machine learning for data-driven discovery in solid earth geoscience

    Bergen, K.J., Johnson, P.A., de Hoop, M.V., Beroza, G.C., 2019. Machine learning for data-driven discovery in solid earth geoscience. Science 363, eaau0323. doi:10.1126/science.aau0323

    work page doi:10.1126/science.aau0323 2019

  3. [4]

    Model reduction and neural networks for parametric pdes

    Bhattacharya, K., Hosseini, B., Kovachki, N.B., Stuart, A.M., 2021. Model reduction and neural networks for parametric pdes. The SMAI journal of computational mathematics 7, 121–157. doi:10.5802/smai-jcm.74

    work page doi:10.5802/smai-jcm.74 2021

  4. [5]

    Approximate distance fields with non- vanishing gradients

    Biswas, A., Shapiro, V., 2004. Approximate distance fields with non- vanishing gradients. Graphical Models 66, 133–159. doi:10.1016/j.gmod. 2004.01.003

    work page doi:10.1016/j.gmod 2004

  5. [6]

    The mathematical theory of finite element methods

    Brenner, S.C., Scott, L.R., Scott, L.R., 2008. The mathematical theory of finite element methods. volume 3. Springer

    2008

  6. [7]

    Physics-informed neural networks for heat transfer problems

    Cai, S., Wang, Z., Wang, S., Perdikaris, P., Karniadakis, G.E., 2021. Physics-informed neural networks for heat transfer problems. Journal of Heat Transfer 143. doi:10.1115/1.4050542

    work page doi:10.1115/1.4050542 2021

  7. [8]

    Dpg loss functions for learning parameter-to-solution maps by neural networks

    Castillo, P.C., Dahmen, W., Gopalakrishnan, J., 2025. Dpg loss functions for learning parameter-to-solution maps by neural networks. arXiv preprint arXiv:2506.18773 doi:10.48550/arXiv.2506.18773

    work page doi:10.48550/arxiv.2506.18773 2025

  8. [9]

    Heterogeneous high frequency seismic radi- ation from complex ruptures

    Cebry, S., McLaskey, G., 2024. Heterogeneous high frequency seismic radi- ation from complex ruptures. Seismica 3. doi:10.26443/seismica.v3i2. 1351

    work page doi:10.26443/seismica.v3i2 2024

  9. [10]

    Creep fronts and complexity in laboratory earth- quake sequences illuminate delayed earthquake triggering

    Cebry, S.B.L., Ke, C.Y., Shreedharan, S., Marone, C., Kammer, D.S., McLaskey, G.C., 2022. Creep fronts and complexity in laboratory earth- quake sequences illuminate delayed earthquake triggering. Nature Commu- nications 13, 6839

    2022

  10. [11]

    Fast computation of the bezout and dixon resultant matrices

    Chionh, E.W., Zhang, M., Goldman, R.N., 2002. Fast computation of the bezout and dixon resultant matrices. Journal of Symbolic Computation 33, 13–29. doi:10.1006/jsco.2001.0462

    work page doi:10.1006/jsco.2001.0462 2002

  11. [12]

    The finite element method for elliptic problems

    Ciarlet, P.G., 2002. The finite element method for elliptic problems. vol- ume 40. Siam. 30

    2002

  12. [13]

    Scientific machine learning through physics–informed neural net- works: Where we are and what’s next

    Cuomo, S., Di Cola, V.S., Giampaolo, F., Rozza, G., Raissi, M., Piccialli, F., 2022. Scientific machine learning through physics–informed neural net- works: Where we are and what’s next. Journal of Scientific Computing 92,

    2022

  13. [14]

    doi:10.1007/s10915-022-01939-z

    work page doi:10.1007/s10915-022-01939-z

  14. [15]

    Data driven approximation of parametrized pdes by reduced basis and neural networks

    Dal Santo, N., Deparis, S., Pegolotti, L., 2020. Data driven approximation of parametrized pdes by reduced basis and neural networks. Journal of Computational Physics 416, 109550. doi:10.1016/j.jcp.2020.109550

    work page doi:10.1016/j.jcp.2020.109550 2020

  15. [16]

    Error estimates for physics informed neural networks approximating the navier-stokes equations

    De Ryck, T., Jagtap, A.D., Mishra, S., 2022. Error estimates for physics informed neural networks approximating the navier-stokes equations. arXiv preprint arXiv:2203.09346 doi:10.48550/arXiv.2203.09346

    work page doi:10.48550/arxiv.2203.09346 2022

  16. [17]

    On the approximation of functions by tanh neural networks

    De Ryck, T., Lanthaler, S., Mishra, S., 2021. On the approximation of functions by tanh neural networks. Neural Networks 143, 732–750. doi:10. 1016/j.neunet.2021.08.015

    2021

  17. [18]

    Erickson, B.A., Jiang, J., Barall, M., Lapusta, N., Dunham, E.M., Harris, R., Abrahams, L.S., Allison, K.L., Ampuero, J.P., Barbot, S., Cattania, C., Elbanna, A., Fialko, Y., Idini, B., Kozdon, J.E., Lambert, V., Liu, Y., Luo, Y., Ma, X., Mckay, M.B., Segall, P., Shi, P., van den Ende, M., Wei, M.,

  18. [19]

    The community code verification exercise for simulating sequences of earthquakes and aseismic slip (SEAS). Seismol. Res. Lett. 91, 874–890. doi:10.1785/0220190248

    work page doi:10.1785/0220190248

  19. [20]

    Theory and practice of finite elements

    Ern, A., Guermond, J.L., 2004. Theory and practice of finite elements. volume 159. Springer. doi:10.1007/978-1-4757-4355-5

    work page doi:10.1007/978-1-4757-4355-5 2004

  20. [21]

    2018updatetotheUS Geological Survey national volcanic threat assessment

    Ewert, J.W., Diefenbach, A.K., Ramsey, D.W., 2018. 2018updatetotheUS Geological Survey national volcanic threat assessment. U.S. Geological Sur- vey Scientific Investigations Report 2018-5140. doi:10.3133/sir20185140

    work page doi:10.3133/sir20185140 2018

  21. [22]

    A tutorial on the adjoint method for inverse problems

    Givoli, D., 2021. A tutorial on the adjoint method for inverse problems. Computer Methods in Applied Mechanics and Engineering 380, 113810. doi:10.1016/j.cma.2021.113810

    work page doi:10.1016/j.cma.2021.113810 2021

  22. [23]

    Transfer learning enhanced physics informed neural network for phase-field model- ing of fracture

    Goswami, S., Anitescu, C., Chakraborty, S., Rabczuk, T., 2020. Transfer learning enhanced physics informed neural network for phase-field model- ing of fracture. Theoretical and Applied Fracture Mechanics 106, 102447. doi:10.1016/j.tafmec.2019.102447

    work page doi:10.1016/j.tafmec.2019.102447 2020

  23. [24]

    A physics-informed deep learning framework for inversion and surrogate mod- eling in solid mechanics

    Haghighat, E., Raissi, M., Moure, A., Gomez, H., Juanes, R., 2021. A physics-informed deep learning framework for inversion and surrogate mod- eling in solid mechanics. Computer Methods in Applied Mechanics and Engineering 379, 113741. doi:10.1016/j.cma.2021.113741

    work page doi:10.1016/j.cma.2021.113741 2021

  24. [25]

    The SCEC/USGS dynamic earthquake rupture code verification exercise

    Harris, R.A., Barall, M., Archuleta, R., Dunham, E.M., Aagaard, B., Am- puero, J.P., Bhat, H., Cruz-Atienza, V., Dalguer, L., Dawson, P., Day, S., Duan, B., Ely, G., Kaneko, Y., Kase, Y., Lapusta, N., Liu, Y., Ma, S., 31 Oglesby, D., Olsen, K., Pitarka, A., Song, S., Templeton, E., 2009. The SCEC/USGS dynamic earthquake rupture code verification exercise....

    work page doi:10.1785/gssrl.80.1.119 2009

  25. [26]

    US geological survey earthquake hazards program decadal science strategy, 2024–33

    Hayes, G.P., Barnhart, W.D., Blanpied, M.L., Davis, L.A., Earle, P.S., Field, N., Franks, J.M., Given, D.D., Gold, R.D., Goulet, C.A., et al., 2024. US geological survey earthquake hazards program decadal science strategy, 2024–33. US Geological Survey Circular 1544. doi:10.3133/cir1544

    work page doi:10.3133/cir1544 2024

  26. [27]

    Fundamentals of computer aided geometric design

    Hoschek, J., Lasser, D., 1993. Fundamentals of computer aided geometric design. AK Peters, Ltd

    1993

  27. [28]

    Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems

    Jagtap, A.D., Kharazmi, E., Karniadakis, G.E., 2020. Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems. Computer Methods in Applied Mechanics and Engineering 365, 113028. doi:10.1016/j.cma. 2020.113028

    work page doi:10.1016/j.cma 2020

  28. [29]

    Nsfnets (navier- stokes flow nets): Physics-informed neural networks for the incompressible navier-stokes equations

    Jin, X., Cai, S., Li, H., Karniadakis, G.E., 2021. Nsfnets (navier- stokes flow nets): Physics-informed neural networks for the incompressible navier-stokes equations. Journal of Computational Physics 426, 109951. doi:https://doi.org/10.1016/j.jcp.2020.109951

    work page doi:10.1016/j.jcp.2020.109951 2021

  29. [30]

    Physics-informed machine learning

    Karniadakis, G.E., Kevrekidis, I.G., Lu, L., Perdikaris, P., Wang, S., Yang, L., 2021. Physics-informed machine learning. Nature Reviews Physics 3, 422–440. doi:10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  30. [31]

    Numerical methods for inverse problems

    Kern, M., 2016. Numerical methods for inverse problems. John Wiley & Sons

    2016

  31. [32]

    Variational physics- informed neural networks for solving partial differential equations

    Kharazmi, E., Zhang, Z., Karniadakis, G.E., 2019. Variational physics- informed neural networks for solving partial differential equations. arXiv preprint arXiv:1912.00873 doi:10.48550/arXiv.1912.00873

    work page doi:10.48550/arxiv.1912.00873 2019

  32. [33]

    hp-vpinns: Variational physics-informed neural networks with domain decomposition

    Kharazmi, E., Zhang, Z., Karniadakis, G.E., 2021. hp-vpinns: Variational physics-informed neural networks with domain decomposition. Computer Methods in Applied Mechanics and Engineering 374, 113547. doi:10.1016/ j.cma.2020.113547

    work page arXiv 2021

  33. [34]

    Deep Learning in Computational Mechanics

    Kollmannsberger, S., D’Angella, D., Jokeit, M., Herrmann, L., et al., 2021. Deep Learning in Computational Mechanics. Springer

    2021

  34. [35]

    The mathematics of artificial intelligence

    Kutyniok, G., 2022. The mathematics of artificial intelligence. arXiv preprint arXiv:2203.08890 doi:10.48550/arXiv.2203.08890

    work page doi:10.48550/arxiv.2203.08890 2022

  35. [36]

    Artificial neural networks for solving ordinary and partial differential equations

    Lagaris, I.E., Likas, A., Fotiadis, D.I., 1998. Artificial neural networks for solving ordinary and partial differential equations. IEEE transactions on neural networks 9, 987–1000. doi:10.1109/72.712178. 32

    work page doi:10.1109/72.712178 1998

  36. [37]

    Neural-network meth- ods for boundary value problems with irregular boundaries

    Lagaris, I.E., Likas, A.C., Papageorgiou, D.G., 2000. Neural-network meth- ods for boundary value problems with irregular boundaries. IEEE Trans- actions on Neural Networks 11, 1041–1049. doi:10.1109/72.870037

    work page doi:10.1109/72.870037 2000

  37. [38]

    Finite difference methods for ordinary and partial differential equations: steady-state and time-dependent problems

    LeVeque, R.J., 2007. Finite difference methods for ordinary and partial differential equations: steady-state and time-dependent problems. SIAM

    2007

  38. [39]

    Heliyon9(4), 14534 (2023) https://doi.org/10.1016/j

    Lundquist, T., Nordström, J., 2014. The sbp-sat technique for initial value problems. Journal of Computational Physics 270, 86–104. doi:10.1016/j. jcp.2014.03.048

    work page doi:10.1016/j 2014

  39. [40]

    Physics-informed neural networks for high-speed flows

    Mao, Z., Jagtap, A.D., Karniadakis, G.E., 2020. Physics-informed neural networks for high-speed flows. Computer Methods in Applied Mechanics and Engineering 360, 112789. doi:10.1016/j.cma.2019.112789

    work page doi:10.1016/j.cma.2019.112789 2020

  40. [41]

    Near-optimal sketchy naturalgradientsforphysics-informedneuralnetworks, in: Forty-secondIn- ternational Conference on Machine Learning

    Mckay, M.B., Kaur, A., Greif, C., Wetton, B., 2025. Near-optimal sketchy naturalgradientsforphysics-informedneuralnetworks, in: Forty-secondIn- ternational Conference on Machine Learning. URL:https://openreview. net/forum?id=bKsZomnmqn

    2025

  41. [42]

    Earthquake initiation from laboratory observations and implications for foreshocks

    McLaskey, G.C., 2019. Earthquake initiation from laboratory observations and implications for foreshocks. Journal of Geophysical Research: Solid Earth 124, 12882–12904

    2019

  42. [43]

    Slow and fast ruptures on a labo- ratory fault controlled by loading characteristics

    Mclaskey, G.C., Yamashita, F., 2017. Slow and fast ruptures on a labo- ratory fault controlled by loading characteristics. Journal of Geophysical Research: Solid Earth 122, 3719–3738

    2017

  43. [44]

    Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for pdes

    Mishra, S., Molinaro, R., 2022a. Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for pdes. IMA Journal of Numerical Analysis 42, 981–1022. doi:10.1093/imanum/drab032

    work page doi:10.1093/imanum/drab032

  44. [45]

    Estimates on the generalization error of physics-informed neural networks for approximating pdes

    Mishra, S., Molinaro, R., 2022b. Estimates on the generalization error of physics-informed neural networks for approximating pdes. IMA Journal of Numerical Analysis doi:10.1093/imanum/drab093

    work page doi:10.1093/imanum/drab093

  45. [46]

    Achieving high accuracy with pinns via energy natural gradients

    Müller, J., Zeinhofer, M., 2023. Achieving high accuracy with pinns via energy natural gradients. arXiv preprint arXiv:2302.13163 doi:10.48550/ arXiv.2302.13163

    work page arXiv 2023

  46. [47]

    Summation-by-parts in time

    Nordström, J., Lundquist, T., 2013. Summation-by-parts in time. Journal of Computational Physics 251, 487–499. doi:10.1016/j.jcp.2013.05.042

    work page doi:10.1016/j.jcp.2013.05.042 2013

  47. [48]

    Raissi, P

    Raissi, M., Perdikaris, P., Karniadakis, G.E., 2019. Physics-informed neu- ral networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Com- putational physics 378, 686–707. doi:10.1016/j.jcp.2018.10.045. 33

    work page doi:10.1016/j.jcp.2018.10.045 2019

  48. [49]

    Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations

    Raissi, M., Yazdani, A., Karniadakis, G.E., 2020. Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations. Science 367, 1026–1030

    2020

  49. [50]

    Verification and Validation in Computational Science and Engineering

    Roache, P., 1998. Verification and Validation in Computational Science and Engineering. 1 ed., Hermosa Publishers, Albuquerque, NM

    1998

  50. [51]

    Physics-informed deep learning of rate- and-state fault friction

    Rucker, C., Erickson, B.A., 2024. Physics-informed deep learning of rate- and-state fault friction. Computer Methods in Applied Mechanics and Engineering 430, 117211. doi:10.1016/j.cma.2024.117211

    work page doi:10.1016/j.cma.2024.117211 2024

  51. [52]

    R-functions in boundary value problems in mechanics

    Rvachev, V.L., Sheiko, T.I., 1995. R-functions in boundary value problems in mechanics. Applied Mechanics Reviews 48, 151–188. URL:https://doi.org/10.1115/1.3005099, doi:10.1115/1.3005099, arXiv:https://asmedigitalcollection.asme.org/appliedmechanicsreviews/article-pdf/48/4/151/5437257/151_- 1.pdf

    work page doi:10.1115/1.3005099 1995

  52. [53]

    On complete- ness of rfm solution structures

    Rvachev, V.L., Sheiko, T.I., Shapiro, V., Tsukanov, I., 2000. On complete- ness of rfm solution structures. Computational Mechanics 25, 305–317. doi:10.1007/s004660050479

    work page doi:10.1007/s004660050479 2000

  53. [54]

    Transfinite interpolation over implicitly defined sets

    Rvachev, V.L., Sheiko, T.I., Shapiro, V., Tsukanov, I., 2001. Transfinite interpolation over implicitly defined sets. Computer aided geometric design 18, 195–220. doi:10.1016/S0167-8396(01)00015-2

    work page doi:10.1016/s0167-8396(01)00015-2 2001

  54. [55]

    Physics-Informed Neural Net- works for Solving Contact Problems in Three Dimensions

    Sahin, T., Wolff, D., Popp, A., 2025. Physics-Informed Neural Net- works for Solving Contact Problems in Three Dimensions. Springer Na- ture Switzerland, Cham. pp. 419–431. URL:https://doi.org/10.1007/ 978-3-031-93213-7_33, doi:10.1007/978-3-031-93213-7_33

    work page doi:10.1007/978-3-031-93213-7_33 2025

  55. [56]

    Implicit repre- sentation of parametric curves and surfaces

    Sederberg, T.W., Anderson, D.C., Goldman, R.N., 1984. Implicit repre- sentation of parametric curves and surfaces. Computer Vision, Graphics, and Image Processing 28, 72–84. doi:10.1016/0734-189X(84)90140-3

    work page doi:10.1016/0734-189x(84)90140-3 1984

  56. [57]

    Theory of R-functions and applications: A primer

    Shapiro, V., 1991. Theory of R-functions and applications: A primer. Technical Report. Cornell University. doi:10.1201/9781315154411

    work page doi:10.1201/9781315154411 1991

  57. [58]

    Semi-analytic geometry with r-functions

    Shapiro, V., 2007. Semi-analytic geometry with r-functions. ACTA numer- ica 16, 239–303

    2007

  58. [59]

    Implicit functions with guaranteed dif- ferential properties, in: Proceedings of the fifth ACM symposium on Solid modeling and applications, pp

    Shapiro, V., Tsukanov, I., 1999. Implicit functions with guaranteed dif- ferential properties, in: Proceedings of the fifth ACM symposium on Solid modeling and applications, pp. 258–269

    1999

  59. [60]

    Physics-informed neural networks with hard and soft boundary conditions for linear free surface waves

    Sheikholeslami, M., Salehi, S., Mao, W., Eslamdoost, A., Nilsson, H., 2025. Physics-informed neural networks with hard and soft boundary conditions for linear free surface waves. Physics of fluids 37. doi:10.1063/5.0277421

    work page doi:10.1063/5.0277421 2025

  60. [61]

    Atwo-dimensionalinterpolationfunctionforirregularly- spaced data, in: Proceedings of the 1968 23rd ACM national conference, pp

    Shepard, D., 1968. Atwo-dimensionalinterpolationfunctionforirregularly- spaced data, in: Proceedings of the 1968 23rd ACM national conference, pp. 517–524. doi:10.1145/800186.810616. 34

    work page doi:10.1145/800186.810616 1968

  61. [62]

    On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type pdes

    Shin, YeonjongDarbon, J., Em Karniadakis, G., 2020. On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type pdes. Communications in Computational Physics 28, 2042–

    2020

  62. [63]

    doi:10.4208/cicp.OA-2020-0193

    work page doi:10.4208/cicp.oa-2020-0193 2020

  63. [64]

    Error estimates of resid- ual minimization using neural networks for linear pdes

    Shin, Y., Zhang, Z., Karniadakis, G.E., 2023. Error estimates of resid- ual minimization using neural networks for linear pdes. Journal of Ma- chine Learning for Modeling and Computing 4, 73–101. doi:10.1615/ JMachLearnModelComput.2023050411

    2023

  64. [65]

    Pinn-fem: A hybrid approach for enforcing dirichlet boundary conditions in physics-informed neural net- works

    Sobh, N., Gladstone, R.J., Meidani, H., 2025. Pinn-fem: A hybrid approach for enforcing dirichlet boundary conditions in physics-informed neural net- works. arXiv preprint arXiv:2501.07765 doi:10.48550/arXiv.2501.07765

    work page doi:10.48550/arxiv.2501.07765 2025

  65. [66]

    Hard- constraining neumann boundary conditions in physics-informed neural net- works via fourier feature embeddings

    Straub, C., Brendel, P., Medvedev, V., Rosskopf, A., 2025. Hard- constraining neumann boundary conditions in physics-informed neural net- works via fourier feature embeddings. arXiv preprint arXiv:2504.01093 doi:10.48550/arXiv.2504.01093

    work page doi:10.48550/arxiv.2504.01093 2025

  66. [67]

    Exact imposition of boundary con- ditions with distance functions in physics-informed deep neural networks

    Sukumar, N., Srivastava, A., 2022. Exact imposition of boundary con- ditions with distance functions in physics-informed deep neural networks. Computer Methods in Applied Mechanics and Engineering 389, 114333. doi:10.1016/j.cma.2021.114333

    work page doi:10.1016/j.cma.2021.114333 2022

  67. [68]

    Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data

    Sun, L., Gao, H., Pan, S., Wang, J.X., 2020. Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data. Computer Methods in Applied Mechanics and Engineering 361, 112732. doi:10.1016/j.cma.2019.112732

    work page doi:10.1016/j.cma.2019.112732 2020

  68. [69]

    Algebraic dis- tance estimations for enriched isogeometric analysis

    Upreti, K., Song, T., Tambat, A., Subbarayan, G., 2014. Algebraic dis- tance estimations for enriched isogeometric analysis. Computer methods in applied mechanics and engineering 280, 28–56. doi:10.1016/j.cma.2014. 07.012

    work page doi:10.1016/j.cma.2014 2014

  69. [70]

    Advanced National Seismic System—Current status, development opportunities, and priorities for 2017–2027

    US Geological Survey, 2017. Advanced National Seismic System—Current status, development opportunities, and priorities for 2017–2027. U.S. Geo- logical Survey Circular 1429. doi:10.3133/cir1429

    work page doi:10.3133/cir1429 2017

  70. [71]

    Multi-scale deep neural network (mscalednn) methods for oscillatory stokes flows in complex domains

    Wang, BoZhang, W., Cai, W., 2020. Multi-scale deep neural network (mscalednn) methods for oscillatory stokes flows in complex domains. Com- munications in Computational Physics 28, 2139–2157. doi:10.4208/cicp. OA-2020-0192

    work page doi:10.4208/cicp 2020

  71. [72]

    The deep ritz method: a deep learning-based numerical algorithm for solving variational problems

    Yu, B., et al., 2018. The deep ritz method: a deep learning-based numerical algorithm for solving variational problems. Communications in Mathemat- ics and Statistics 6, 1–12. doi:10.1007/s40304-018-0127-z

    work page doi:10.1007/s40304-018-0127-z 2018

  72. [73]

    Machine learning for predicting battery capacity for electric vehicles

    Zhao, J., Ling, H., Liu, J., Wang, J., Burke, A.F., Lian, Y., 2023. Machine learning for predicting battery capacity for electric vehicles. eTransporta- tion 15, 100214. doi:10.1016/j.etran.2022.100214. 35

    work page doi:10.1016/j.etran.2022.100214 2023