Pith. sign in

REVIEW 2 major objections 5 minor 50 references

Mass weighting algorithm optimizes Fourier-based physics-informed neural network in adhesive contact mechanics

T0 review · 2 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read A mass-weighting spectral preconditioner lets energy-minimizing Fourier PINNs reach machine-zero residual on adhesive contact problems and match GFMD displacement and stress fields.

desk verdict Clean, well-ablated spectral fix that makes energy-based Fourier PINNs actually converge on adhesive half-space contact; 1-D evidence is solid, free parameters and missing code are the real soft spots. read the letter →

arxiv 2607.04288 v1 pith:MQHRTRNH submitted 2026-07-05 cond-mat.soft

Yunong Zhou , Kaifeng Huang , Chaofan Du , Yang Xu , Hengxu Song This is my paper · ORCID
classification cond-mat.soft
keywords contactmechanicsadhesionPINNmassweightingspectralpreconditioningFourierroughsurfacesenergyminimization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Physics-informed neural networks that minimize the total potential energy of an elastic contact problem stall because the elastic stiffness grows linearly with wave number: short-wavelength modes dominate the gradients and the macroscopic shape never settles. This paper shows that reweighting those gradients in Fourier space with a mass-weighting function (amplifying low-q modes and low-pass filtering high-q noise) removes the imbalance. On adhesive line-contact benchmarks the preconditioned network reaches machine-zero residual within a few hundred Adam steps and produces displacement and contact-stress fields that agree with Green’s-function molecular dynamics to within a fraction of a percent, for both smooth Hertz indenters and fractal rough surfaces spanning tension to compression. The method works on a plain uniform grid, needs no Green’s-function assembly or quadrature, and is written entirely as scalar energy minimization, so the same spectral idea transfers directly to two-dimensional rough surfaces.

What carries the argument

The mass-weighting spectral preconditioner G(q)=w(q)f(q), where w(q) reweights Fourier modes of the displacement gradient to amplify low-wavenumber contributions and f(q) is a built-in low-pass filter that suppresses sub-grid noise; the reweighted gradient is inverse-transformed and used for ordinary back-propagation.

What would settle it

Train the identical network and energy functional on the same Hertz or rough-surface cases with mass weighting disabled (or with a deliberately mismatched G(q)); if residual loss still reaches machine zero and the stress field remains oscillation-free and GFMD-accurate, the claimed necessity of spectral rebalancing is false.

Watch

Extended reading notes

Core claim

Spectral stiffness imbalance is the dominant obstacle to training energy-minimizing Fourier PINNs for elastic contact; a mass-weighting preconditioner applied to the displacement gradient in Fourier space before back-propagation eliminates that imbalance, driving residual loss to machine zero within roughly 400 Adam iterations and recovering displacement and stress fields that match GFMD reference solutions quantitatively for both smooth and multi-scale rough adhesive contacts.

Load-bearing premise

The preconditioner is assumed to work because elastic stiffness dominates the gradient spectrum; if short-range adhesive forces instead dominate the high-wavenumber gradients, the fixed reweighting may fail to rebalance the modes.

Editorial extensions

If this is right

  • Energy-minimizing PINNs become a practical mesh-free alternative to GFMD or BEM for one-dimensional adhesive line contact once spectral preconditioning is used.
  • The same Fourier elastic energy and wave-number-only preconditioner transfer without structural change to two-dimensional rough surfaces.
  • Smooth real-space interaction potentials other than Morse (e.g., Lennard-Jones adhesion or rate-dependent friction) can be substituted without redesigning the spectral reweighting.
  • Training dynamics no longer require manual balancing of multi-term loss weights; a single scalar energy suffices once gradients are spectrally rebalanced.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same spectral-reweighting idea is likely to help any energy-based PINN whose linear operator has a strongly growing Fourier symbol, not only contact elasticity.
  • If the low-pass cutoff is tied to the interaction range, the method may automatically adapt to potentials of different interaction widths without retuning.
  • Failure modes on three-dimensional or finite-body geometries would most probably appear first as residual high-q oscillations in the stress field rather than as slow macroscopic convergence.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript introduces a spectral mass-weighting (MW) preconditioner for energy-minimizing Fourier PINNs applied to one-dimensional adhesive elastic contact. The elastic kernel qE*/2 grows linearly with wavenumber and causes high-q modes to dominate Adam updates, stalling macroscopic convergence. Before back-propagation the displacement gradient is Fourier-transformed, reweighted by G(q)=w(q)f(q) (Eqs. 12–14) that amplifies low-q modes and applies a low-pass filter, then inverse-transformed. On Hertz and fractal (H=0.5) Morse-adhesive line contacts the MW-PINN reaches machine-zero residual within ~400 Adam steps (vs. stalling three orders higher without MW) and yields displacement/stress fields whose NRMSE relative to independent GFMD is at most ~10^{-3} across six pressure cases (Table 1).

Significance. If the reported ablations and GFMD agreement hold, the work supplies a practical, energy-only training device that removes a concrete spectral-stiffness obstacle for Fourier PINNs in contact mechanics. Strengths that should be credited are the direct MW-on/off loss and fastest-mode diagnostics (Figs. 3–5), the quantitative RMSE/NRMSE table against an external GFMD reference that does not depend on the PINN or the MW parameters (Table 1), and the clean energy formulation that needs no PDE residual or boundary-penalty balancing. The method is restricted to 1-D line contact with Morse adhesion, but within that scope the evidence is concrete and the free parameters of G(q) are stated.

major comments (2)
  1. End of §2.2 and the free-parameter list: the central claim that MW is “necessary” for stable convergence rests on a fixed G(q) with α=0.5, ξ, qc, β and clamps that are never varied. A short sensitivity study (or at least a statement of the ranges that still reach machine-zero loss) is needed to show that the acceleration is not an artifact of one hand-tuned setting; without it the necessity claim is only partially supported.
  2. Abstract and §4: the assertion that “extension to two-dimensional rough surfaces is direct” is stronger than the evidence. The paper itself notes that effectiveness for other potentials or 3-D geometries “remains to be investigated.” Soften the language to a conjecture or supply a minimal 2-D demonstration; otherwise the claim over-reaches the 1-D Morse benchmarks that actually support the quantitative results.
minor comments (5)
  1. Abstract and Fig. 3 caption: “machine-zero residual loss” should be quantified (e.g., floating-point floor or a concrete threshold) so that the three-order improvement is unambiguous.
  2. Eq. (12): the low-q floor k0=ξ(E*qmax/2) and the clamps [wmin,wmax] are introduced without numerical values used in the runs; list them for reproducibility.
  3. Fig. 3 y-label and caption contain the typo “Resccaled”; correct to “Rescaled.”
  4. Data-availability statement: replace the placeholder “can be found here” with an actual repository link or DOI before publication.
  5. Several sentences lack spaces after commas or periods (e.g., abstract “imbalance,that is”); a light copy-edit pass would improve readability.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: energy functional and GFMD benchmarks are independent of the MW training device; self-citations supply background only.

full rationale

The derivation chain is: classical half-space elastic energy (Fourier kernel qE*/2, Eq. 5) + external work + Morse interaction (Eqs. 6–9) define a scalar potential Π[u]; a Fourier-feature NN parameterizes u_θ; Adam minimizes Π after a spectral reweighting G(q)=w(q)f(q) of the displacement gradient (Eqs. 12–14). Ablations (Figs. 3–5) show that without MW the loss stalls and stress is noisy; with MW the residual reaches machine zero and fields match GFMD. GFMD solutions are obtained by an independent numerical method (fast Fourier Green’s-function iteration) that does not depend on the PINN architecture, the mass-weighting parameters (α=0.5, ξ, qc, β), or the network weights. The MW function itself is a pure training preconditioner, not part of the physics being solved or predicted. Self-citations ([36] for the mass-weighting idea, [44,45] for prior PINN contact work, [30] for field theory) appear only as background or motivation; none of them supplies the residual-loss curves, the NRMSE table, or the claim that MW is necessary for stable convergence. No parameter is fitted to the target residual or NRMSE and then re-presented as a prediction; no uniqueness theorem is imported to force the form of G(q); the energy functional is not defined in terms of the MW weights. The paper’s own caveat (end of §2.2) that effectiveness for other potentials or 3-D geometries remains open is an honest scope limitation, not a circular step. Consequently the central quantitative claims rest on external numerical evidence and internal controlled ablations, not on self-referential construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The physics (Fourier elastic energy of a half-space, Morse adhesion, energy minimization) is standard. The paper's contribution is the engineered preconditioner G(q)=w(q)f(q) and the demonstration that it stabilizes PINN training. Several hand-chosen hyperparameters of that preconditioner and of the network are free parameters the central numerical claims depend on; they are not derived from first principles.

free parameters (6)
  • mass-weighting exponent α = 0.5
    Fixed to 0.5 'in this study' (Eq. 12); controls how strongly low-q modes are amplified. No derivation or sweep reported.
  • low-q stiffness floor factor ξ = ≪1 (unspecified)
    k0=ξ(E* qmax/2) with ξ≪1; sets the floor that prevents infinite gain at q→0. Value not numerically specified beyond ≪1.
  • spectral filter cutoff qc and roll-off β
    qc set 'proportional to the reciprocal of the interaction range'; β controls steepness (Eq. 13). Exact constants not given.
  • weight clamps [wmin, wmax]
    w(q) is clamped to prevent extreme gains; bounds not numerically reported.
  • warm-up steps Nwarm = 100–200
    Preconditioner activated only after 100–200 plain-gradient steps; chosen for stability, not derived.
  • Adam learning rate and network width/depth = lr=1e-4
    lr=10^{-4}; nl hidden layers, nh neurons, Fourier-feature scale s≈πL/nx. Architecture details partially schematic.
assumptions (5)
  • domain assumption Elastic energy of a semi-infinite isotropic half-space is Uel=(L/2) Σ_q (q E*/2)|ũ(q)|² (Eq. 5).
    Standard Fourier representation of the half-space Green's function (Persson, Prodanov et al.); assumed exact for the continuum model.
  • domain assumption Adhesive interaction is fully described by a Morse potential γ(g) with range ρ≪L (Eq. 7).
    Modeling choice; other potentials (LJ, Dugdale) are left for future work.
  • domain assumption The contact problem is equivalent to unconstrained minimization of the scalar total potential Π[u] (Eq. 10); no explicit PDE residual or BC penalty is required.
    Energy-based PINN formulation; valid for frictionless normal contact under the stated kinematics.
  • ad hoc to paper Gradient spectrum is dominated by the elastic stiffness contribution, so reweighting by a function of the elastic kernel rebalances training.
    Stated as an assumption at the end of §2.2; effectiveness for other potentials/3-D 'remains to be investigated.'
  • domain assumption Plane-strain line contact on a uniform 1-D grid with periodic Fourier representation is sufficient to demonstrate the method.
    Explicit scope restriction in §2.2; 2-D claimed to be direct but not shown.
invented entities (1)
  • Mass-weighting spectral preconditioner G(q)=w(q)f(q)
    purpose: Reweight ∂Π/∂u in Fourier space before back-propagation so low-q modes receive amplified gradients and sub-grid noise is suppressed.
    Core algorithmic invention of the paper (Eqs. 12–14). Independent evidence is the MW-on vs MW-off ablation and GFMD match; no external theoretical derivation of the specific functional form.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mass weighting algorithm optimizes Fourier-based physics-informed neural network in adhesive contact mechanics." pith.science (2026). https://pith.science/paper/MQHRTRNH

@misc{pith2026260704288,
  author       = {Pith},
  title        = {Pith review of: Mass weighting algorithm optimizes Fourier-based physics-informed neural network in adhesive contact mechanics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MQHRTRNH}},
  note         = {Machine review of arXiv:2607.04288}
}
read the original abstract

Physics-informed neural networks (PINNs) for elastic contact mechanics suffer from a spectral stiffness imbalance,that is, the elastic kernel grows linearly with wave number, causing short-wavelength modes to dominate gradient updates and stall convergence of the macroscopic deformation. We introduce a spectral preconditioning strategy that reweights displacement gradients in Fourier space before back-propagation, amplifying low wavenumber components through a mass weighting (MW) function while suppressing sub-grid noise via a built-in low-pass filter. Applied to adhesive line contact problems, the mass weighted PINN reaches machine-zero residual loss within 400 Adam iterations for specified benchmark, whereas the reference benchmark stalls at three orders of magnitude higher loss. The converged displacement and contact stress fields agree quantitatively with Green's function molecular dynamics (GFMD) solutions for both smooth Hertz contact at pressures spanning tension to compression and rough surfaces with roughness covering several decades of wavelength. The method operates directly on a uniform real-space grid, requires no explicit Green's function integration or quadrature rules, and is formulated entirely in terms of minimising a scalar energy function. Extension to two-dimensional rough surfaces is direct, as both the Fourier elastic energy and the spectral preconditioner depend only on the wave-number magnitude.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 30 canonical work pages

  1. [1]

    Mechanics of Materials37(2-3), 275–285 (2005)

    Gao, H., Wang, X., Yao, H., Gorb, S., Arzt, E.: Mechanics of hierarchical adhesion structures of geckos. Mechanics of Materials37(2-3), 275–285 (2005). https://doi.org/10.1016/j.mechmat.2004.03.008

    work page doi:10.1016/j.mechmat.2004.03.008 2005

  2. [2]

    Langmuir25(13), 7486– 7495 (2009)

    Zeng, H., Pesika, N., Tian, Y., Zhao, B., Chen, Y., Tirrell, M., Turner, K.L., Israelachvili, J.N.: Frictional adhesion of patterned surfaces and implications for gecko and biomimetic systems. Langmuir25(13), 7486– 7495 (2009). https://doi.org/10.1021/la900877h Springer Nature 2021 LATEX template Fourier PINN adhesive contact mechanics17

    work page doi:10.1021/la900877h 2009

  3. [3]

    Small18(35) (2022)

    Materzok, T., De Boer, D., Gorb, S., M¨ uller-Plathe, F.: Gecko adhesion on flat and rough surfaces: Simulations with a multi-scale molecular model. Small18(35) (2022). https://doi.org/10.1002/smll.202201674

    work page doi:10.1002/smll.202201674 2022

  4. [4]

    The Journal of Chemical Physics162(7) (2025)

    Persson, B.N.J., Xu, R., Miyashita, N.: Rubber wear: Experiment and theory. The Journal of Chemical Physics162(7) (2025). https://doi.org/ 10.1063/5.0248199

    work page doi:10.1063/5.0248199 2025

  5. [5]

    Tribology Letters73(3) (2025)

    Xu, R., Sheng, W., Zhou, F., Persson, B.N.J.: Rubber wear: History, mechanisms, and perspectives. Tribology Letters73(3) (2025). https: //doi.org/10.1007/s11249-025-02025-9

    work page doi:10.1007/s11249-025-02025-9 2025

  6. [6]

    Physical Review Letters108(24) (2012)

    Dapp, W.B., L¨ ucke, A., Persson, B.N.J., M¨ user, M.H.: Self-affine elas- tic contacts: Percolation and leakage. Physical Review Letters108(24) (2012). https://doi.org/10.1103/physrevlett.108.244301

    work page doi:10.1103/physrevlett.108.244301 2012

  7. [7]

    Physical Review Letters131(15) (2023)

    M¨ uller, C., M¨ user, M.H., Carbone, G., Menga, N.: Significance of elastic coupling for stresses and leakage in frictional contacts. Physical Review Letters131(15) (2023). https://doi.org/10.1103/physrevlett.131.156201

    work page doi:10.1103/physrevlett.131.156201 2023

  8. [8]

    International Journal of Mechanical Sciences284, 109740 (2024)

    Li, H., Yu, C., Dai, Z.: Regimes in the axisymmetric stiction of thin elastic plates. International Journal of Mechanical Sciences284, 109740 (2024). https://doi.org/10.1016/j.ijmecsci.2024.109740

    doi:10.1016/j.ijmecsci.2024.109740 2024

Show all 50 references
  1. [9]

    Tribology Letters19(1), 59–63 (2005)

    McClelland, G.M., Rettner, C.T., Hart, M.W., Carter, K.R., Sanchez, M.I., Best, M.E., Terris, B.D.: Contact mechanics of a flexible imprinter for photocured nanoimprint lithography. Tribology Letters19(1), 59–63 (2005). https://doi.org/10.1007/s11249-005-4265-6

    2005 doi

  2. [10]

    Advanced Functional Mate- rials22(21), 4476–4484 (2012)

    Carlson, A., Wang, S., Elvikis, P., Ferreira, P.M., Huang, Y., Rogers, J.A.: Active, programmable elastomeric surfaces with tunable adhesion for deterministic assembly by transfer printing. Advanced Functional Mate- rials22(21), 4476–4484 (2012). https://doi.org/10.1002/adfm.201201023

    2012 doi

  3. [11]

    In: Journal F¨ ur die Reine und Angewandte Mathematik, vol

    Hertz, H.: On the contact of elastic solids. In: Journal F¨ ur die Reine und Angewandte Mathematik, vol. 92, pp. 156–171 (1882)

  4. [12]

    Proceedings of the Royal Society of London

    Johnson, K.L., Kendall, K., Roberts, A.D.: Surface energy and the con- tact of elastic solids. Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences324(1558), 301–313 (1971). https: //doi.org/10.1098/rspa.1971.0141

    1971 doi

  5. [13]

    Journal of Colloid and Interface Science 53(2), 314–326 (1975)

    Derjaguin, B.V., Muller, V.M., Toporov, Y.P.: Effect of contact deforma- tions on the adhesion of particles. Journal of Colloid and Interface Science 53(2), 314–326 (1975). https://doi.org/10.1016/0021-9797(75)90018-1

    1975 doi

  6. [14]

    Journal of Colloid and Interface Science150(1), 243–269 (1992)

    Maugis, D.: Adhesion of spheres: The jkr-dmt transition using a dugdale Springer Nature 2021 LATEX template 18Fourier PINN adhesive contact mechanics model. Journal of Colloid and Interface Science150(1), 243–269 (1992). https://doi.org/10.1016/0021-9797(92)90285-t

    2021 doi

  7. [16]

    The Journal of Adhesion 99(6), 972–987 (2022)

    Ciavarella, M., Wang, Q.-A., Li, Q.: Maugis-tabor parameter dependence of pull-off in viscoelastic line hertzian contacts. The Journal of Adhesion 99(6), 972–987 (2022). https://doi.org/10.1080/00218464.2022.2066998

    2022 doi

  8. [17]

    Proceedings of the Royal Society of London

    Greenwood, J.A., Williamson, J.B.P.: Contact of nominally flat surfaces. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences295(1442), 300–319 (1966). https://doi.org/10.1098/ rspa.1966.0242

    1966

  9. [18]

    The Journal of Chemical Physics115(8), 3840–3861 (2001)

    Persson, B.N.J.: Theory of rubber friction and contact mechanics. The Journal of Chemical Physics115(8), 3840–3861 (2001). https://doi.org/ 10.1063/1.1388626

    2001 doi

  10. [19]

    The European Physical Journal E8(4), 385–401 (2002)

    Persson, B.N.J.: Adhesion between an elastic body and a randomly rough hard surface. The European Physical Journal E8(4), 385–401 (2002). https://doi.org/10.1140/epje/i2002-10025-1

    2002 doi

  11. [20]

    Physical Review Letters99(12) (2007)

    Persson, B.N.J.: Relation between interfacial separation and load: A gen- eral theory of contact mechanics. Physical Review Letters99(12) (2007). https://doi.org/10.1103/physrevlett.99.125502

    2007 doi

  12. [21]

    Physical Review E87(6) (2013)

    Pastewka, L., Prodanov, N., Lorenz, B., M¨ user, M.H., Robbins, M.O., Persson, B.N.J.: Finite-size scaling in the interfacial stiffness of rough elas- tic contacts. Physical Review E87(6) (2013). https://doi.org/10.1103/ physreve.87.062809

    2013

  13. [22]

    Physical Review E70(2) (2004)

    Hyun, S., Pei, L., Molinari, J.-F., Robbins, M.O.: Finite-element analysis of contact between elastic self-affine surfaces. Physical Review E70(2) (2004). https://doi.org/10.1103/physreve.70.026117

    2004 doi

  14. [23]

    Journal of Physics: Condensed Matter21(1), 015003 (2008)

    Lorenz, B., Persson, B.N.J.: Interfacial separation between elastic solids with randomly rough surfaces: comparison of experiment with theory. Journal of Physics: Condensed Matter21(1), 015003 (2008). https://doi. org/10.1088/0953-8984/21/1/015003

    2008 doi

  15. [24]

    The Journal of Chemical Physics120(18), 8779–8793 (2004)

    Persson, B.N.J., Albohr, O., Creton, C., Peveri, V.: Contact area between a viscoelastic solid and a hard, randomly rough, substrate. The Journal of Chemical Physics120(18), 8779–8793 (2004). https://doi.org/10.1063/1. 1697376 Springer Nature 2021 LATEX template Fourier PINN a...

    2004 doi

  16. [25]

    The Journal of Chemical Physics148(14) (2018)

    Persson, B.N.J.: The dependency of adhesion and friction on electrostatic attraction. The Journal of Chemical Physics148(14) (2018). https://doi. org/10.1063/1.5024038

    2018 doi

  17. [26]

    Journal of Physics: Condensed Matter33(43), 435001 (2021)

    Persson, B.N.J.: General theory of electroadhesion. Journal of Physics: Condensed Matter33(43), 435001 (2021). https://doi.org/10.1088/ 1361-648x/abe797

    2021

  18. [27]

    Tribology Letters 70(2) (2022)

    Persson, B.N.J.: Fluid leakage in static rubber seals. Tribology Letters 70(2) (2022). https://doi.org/10.1007/s11249-022-01573-8

    2022 doi

  19. [28]

    Tribology International214, 111352 (2026)

    Xu, R., Gil, L., Singer, J., Gontard, L., Leverd, W., Persson, B.N.J.: Leakage at interfaces: A comprehensive study based on persson contact mechanics theory. Tribology International214, 111352 (2026). https:// doi.org/10.1016/j.triboint.2025.111352

    2026 doi

  20. [29]

    Physical Review Letters100(5) (2008)

    M¨ user, M.H.: Rigorous field-theoretical approach to the contact mechanics of rough elastic solids. Physical Review Letters100(5) (2008). https:// doi.org/10.1103/physrevlett.100.055504

    2008 doi

  21. [30]

    Tribology International220, 111894 (2026)

    Zhou, Y., Song, H., Zhang, Z., Xu, Y.: Field-theoretical approach to esti- mate mean gap and gap distribution in randomly rough surface contact mechanics. Tribology International220, 111894 (2026). https://doi.org/ 10.1016/j.triboint.2026.111894

    2026 doi

  22. [31]

    Computational Mechan- ics32(1-2), 141–141 (2003)

    Wriggers, P.: Computational contact mechanics. Computational Mechan- ics32(1-2), 141–141 (2003). https://doi.org/10.1007/s00466-003-0472-x

    2003 doi

  23. [32]

    Phys- ical Mesomechanics17(4), 321–333 (2014)

    Putignano, C., Carbone, G.: A review of boundary elements method- ologies for elastic and viscoelastic rough contact mechanics. Phys- ical Mesomechanics17(4), 321–333 (2014). https://doi.org/10.1134/ s1029959914040092

    2014

  24. [33]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 470(2169), 20140392 (2014)

    Menga, N., Putignano, C., Carbone, G., Demelio, G.P.: The sliding con- tact of a rigid wavy surface with a viscoelastic half-space. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 470(2169), 20140392 (2014). https://doi.org/10.1098/rspa.2014.0392

    2014 doi

  25. [34]

    Physical Review E103(5) (2021)

    Monti, J.M., Pastewka, L., Robbins, M.O.: Green’s function method for dynamic contact calculations. Physical Review E103(5) (2021). https: //doi.org/10.1103/physreve.103.053305

    2021 doi

  26. [35]

    Physical Review B74(7) (2006)

    Campa˜ n´ a, C., M¨ user, M.H.: Practical green’s function approach to the simulation of elastic semi-infinite solids. Physical Review B74(7) (2006). https://doi.org/10.1103/physrevb.74.075420 Springer Nature 2021 LATEX template 20Fourier PINN adhesive contact mechanics

    2006 doi

  27. [36]

    Physical Review B 99(14) (2019)

    Zhou, Y., Moseler, M., M¨ user, M.H.: Solution of boundary-element prob- lems using the fast-inertial-relaxation-engine method. Physical Review B 99(14) (2019). https://doi.org/10.1103/physrevb.99.144103

    2019 doi

  28. [37]

    Frontiers in Mechanical Engineering6(2021)

    Kalliorinne, K., Larsson, R., P´ erez-R` afols, F., Liwicki, M., Almqvist, A.: Artificial neural network architecture for prediction of contact mechanical response. Frontiers in Mechanical Engineering6(2021). https://doi.org/ 10.3389/fmech.2020.579825

    2021 doi

  29. [38]

    Journal of Tribology 147(11) (2025)

    Suman, S., Prajapati, D.K.: Predictive modeling of real contact area on rough surfaces using deep artificial neural network. Journal of Tribology 147(11) (2025). https://doi.org/10.1115/1.4068057

    2025 doi

  30. [39]

    Science324(5923), 81–85 (2009)

    Schmidt, M., Lipson, H.: Distilling free-form natural laws from experi- mental data. Science324(5923), 81–85 (2009). https://doi.org/10.1126/ science.1165893

    2009

  31. [40]

    Proceedings of the National Academy of Sciences 104(24), 9943–9948 (2007)

    Bongard, J., Lipson, H.: Automated reverse engineering of nonlinear dynamical systems. Proceedings of the National Academy of Sciences 104(24), 9943–9948 (2007). https://doi.org/10.1073/pnas.0609476104

    2007 doi

  32. [41]

    Pro- ceedings of the National Academy of Sciences113(15), 3932–3937 (2016)

    Brunton, S.L., Proctor, J.L., Kutz, J.N.: Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Pro- ceedings of the National Academy of Sciences113(15), 3932–3937 (2016). https://doi.org/10.1073/pnas.1517384113

    2016 doi

  33. [43]

    Nature Communications12(1) (2021)

    Chen, Z., Liu, Y., Sun, H.: Physics-informed learning of governing equations from scarce data. Nature Communications12(1) (2021). https: //doi.org/10.1038/s41467-021-26434-1

    2021 doi

  34. [44]

    Tribology Letters73(88) (2025)

    Zhou, Y., Song, H.: Physics-informed neural network approach to ran- domly rough surface contact mechanics. Tribology Letters73(88) (2025). https://doi.org/10.1007/s11249-025-02022-y

    2025 doi

  35. [45]

    Tribology International214, 111100 (2026)

    Zhou, Y., Song, H.: Data-driven enhanced rough contact mechanics: Pinn estimation of gap distribution across length scales for partial contacts. Tribology International214, 111100 (2026). https://doi.org/10.1016/j. triboint.2025.111100

    2026 doi

  36. [46]

    Journal of the Mechanics and Physics of Solids210, 106542 (2026)

    Li, H., Li, M., Bai, J., Sharif Khodaei, Z., Aliabadi, M.H.: An energy- based physics-informed neural network framework for efficient mesh-free Springer Nature 2021 LATEX template Fourier PINN adhesive contact mechanics21 homogenisation of architected metamaterials. Journal of...

    2021 doi

  37. [47]

    Computer Methods in Applied Mechanics and Engineering437, 117787 (2025)

    Bai, J., Lin, Z., Wang, Y., Wen, J., Liu, Y., Rabczuk, T., Gu, Y., Feng, X.-Q.: Energy-based physics-informed neural network for frictionless con- tact problems under large deformation. Computer Methods in Applied Mechanics and Engineering437, 117787 (2025). https://doi.org/10...

    2025

  38. [48]

    SIAM Journal on Scientific Computing43(5), 3055–3081 (2021)

    Wang, S., Teng, Y., Perdikaris, P.: Understanding and mitigating gradient flow pathologies in physics-informed neural networks. SIAM Journal on Scientific Computing43(5), 3055–3081 (2021). https://doi.org/10.1137/ 20m1318043

    2021

  39. [49]

    arXiv (2026)

    He, R.Y., Liang, Y., Zhao, H., Zhong, Y.: From Frequency Bias to Spectral Balance: Operator-Aware Preconditioners for PINNs. arXiv (2026). https://doi.org/10.48550/ARXIV.2603.00904. https://arxiv.org/ abs/2603.00904

    2026 doi

  40. [50]

    In: Chaudhuri, K., Salakhutdinov, R

    Rahaman, N., Baratin, A., Arpit, D., Draxler, F., Lin, M., Hamprecht, F., Bengio, Y., Courville, A.: On the spectral bias of neural networks. In: Chaudhuri, K., Salakhutdinov, R. (eds.) Proceedings of the 36th International Conference on Machine Learning. Proceedings of Machin...

    2019

  41. [51]

    In: Larochelle, H., Ranzato, M., Hadsell, R., Bal- can, M.F., Lin, H

    Tancik, M., Srinivasan, P., Mildenhall, B., Fridovich-Keil, S., Ragha- van, N., Singhal, U., Ramamoorthi, R., Barron, J., Ng, R.: Fourier features let networks learn high frequency functions in low dimen- sional domains. In: Larochelle, H., Ranzato, M., Hadsell, R., Bal- can, ...

    2020

  42. [52]

    Tribology Letters53(2), 433–448 (2013)

    Prodanov, N., Dapp, W.B., M¨ user, M.H.: On the contact area and mean gap of rough, elastic contacts: Dimensional analysis, numerical cor- rections, and reference data. Tribology Letters53(2), 433–448 (2013). https://doi.org/10.1007/s11249-013-0282-z

    2013 doi

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.