Pith. sign in

REVIEW 4 major objections 5 minor 23 references

Data-driven model reconstruction for nonlinear wave dynamics

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Sparse regression reconstructs the nonlinear envelope equation governing valley-Hall edge wavepackets in photonic lattices.

desk verdict A useful parameter-identification demonstration for topological photonic edge waves, with an unsupported 'no a priori assumptions' claim in the abstract. read the letter →

arxiv 2411.11556 v1 pith:UBWPNYOT submitted 2024-11-18 physics.optics cs.LGmath-phmath.MPphysics.data-an

Ekaterina Smolina , Lev Smirnov , Daniel Leykam , Franco Nori , Daria Smirnova This is my paper · ORCID
classification physics.opticscs.LGmath-phmath.MPphysics.data-an
keywords sparseregressionvalley-HalldomainwallsphotoniclatticesnonlinearSchrödingerequationself-steepeningself-focusinginterpretablemachinelearningtopologicalphotonics
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

The paper shows that a sparse-regression machine-learning scheme can infer, from beam-propagation data alone, the effective continuum partial differential equation that governs the envelope of valley-Hall edge wavepackets in honeycomb photonic lattices. The reconstructed model reproduces the linear dispersion of the interface states and the dominant nonlinear effects—self-steepening and self-focusing—for both zigzag and bearded domain-wall geometries. The authors argue that this data-driven route bypasses the scale-hierarchy assumptions that constrain conventional asymptotic multi-scale reductions, yielding an interpretable model rather than a black-box predictor.

What carries the argument

The central machinery is the sparse-regression pipeline applied to the discretely sampled envelope A(z,x), with data enriched by sweeping the input beam position across the lattice period to build a smooth continuum function. A large library of candidate terms—differential operators up to third order in x and polynomial nonlinearities up to quintic in the envelope—is assembled, and a sparsity-promoting regression of the lasso type selects the few terms with predictive power. The split-step strategy is also load-bearing: linear coefficients are reconstructed first from low-intensity data, and only the residual (i∂z A − L̂A) is regressed for nonlinear coefficients, which keeps the two regimes from contaminating each other.

What would settle it

Run the same regression on paraxial data from a lattice that knowingly includes a nonlocal or higher-derivative coupling in the envelope dynamics; if the fixed form of Eq. (4) with the regression coefficients fails to predict the full simulation at propagation distances beyond the training window, while a model enriched with the known term succeeds, then the claim that the ansatz is free of a priori limitations would be refuted.

Watch

Extended reading notes

Core claim

On its own terms, the discovery is that a scalar envelope equation of a fixed functional form—Eq. (4), containing first-, second-, and third-order spatial derivatives, cubic and quintic nonlinear terms, and nonlinear group-velocity corrections—is sufficient to describe the slow evolution of edge wavepackets at both zigzag and bearded valley-Hall domain walls in a laser-written waveguide lattice. The coefficients of this equation are determined by sparse regression on data from paraxial simulations: first the linear operator is fixed at low input intensity, then the residual nonlinear correction is fit at higher intensity. The resulting model reproduces the band-structure dispersion near the K+ point and captures the qualitative beam-shaping effects, including a self-steepening asymmetry and self-focusing compression, with coefficients that the authors validate against direct numerical propagation.

Load-bearing premise

The load-bearing premise is that the candidate library—derivatives up to third order and polynomial nonlinearities up to quintic—contains all the terms that actually control the envelope dynamics, so any effect missing from that library will be silently folded into the selected ones.

Editorial extensions

If this is right

  • Edge-wave dynamics can be predicted by integrating a one-dimensional envelope PDE instead of the full two-dimensional paraxial lattice model.
  • The approach reconstructs nonlinear coefficients directly from propagation data, bypassing the integral calculations of asymptotic multi-scale theory.
  • The same regression scheme can be applied to other domain-wall shapes and lattice parameters without redoing the analytical reduction.
  • Higher-order effects such as third-order dispersion and quintic nonlinearity are recovered automatically when they matter at the explored propagation distances.
  • Because the recovered model is an equation with physical terms, it can be used for design and for interpreting which nonlinear mechanism dominates.

Reading between the lines

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

  • A natural extension would be to feed the regression with experimental camera images of the envelope; the main obstacles would be noise and the need for accurate derivative estimation, but the method's structure already accommodates intensity-based data.
  • If the ansatz library were deliberately enlarged (e.g., with nonlocal convolution terms or coupling to a second envelope field), the regression could serve as a diagnostic for whether a single-field local PDE is at all valid for a given interface.
  • The fixed-form equation (4) should be seen as a validation of the physical mechanisms included, not a guarantee that the true microscopic dynamics reduce to it; a poor out-of-sample fit at longer distances would signal missing physics.
  • The split-step linear-then-nonlinear procedure suggests a possible iterative loop: any systematic residual after fitting Eq. (4) could seed a new library, turning the method into an automated model-discovery tool for unknown wave media.
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

4 major / 5 minor

Summary. The manuscript presents an interpretable machine-learning framework for reconstructing an effective one-dimensional continuum equation for the envelope of valley-Hall edge wavepackets in honeycomb photonic lattices. The authors perform full paraxial simulations of zigzag and bearded domain walls for two parameter sets, extract a waveguide-centered envelope by sweeping the input beam position between lattice sites, and use sparse regression to identify coefficients in a library of linear derivative terms (up to third order) and polynomial nonlinear terms (up to quintic). They report that the reconstructed linear operator reproduces the edge-state dispersion and that the nonlinear terms capture self-steepening and self-focusing at higher input intensities. The abstract and conclusion claim that the scheme is 'proven free of the a priori limitations imposed by the underlying hierarchy of scales' of asymptotic methods.

Significance. If the result is correct, the paper would demonstrate a valuable proof-of-concept: sparse regression can produce an interpretable continuum PDE for nonlinear topological photonic edge dynamics without an analytical multiscale reduction. The two-stage treatment of linear and nonlinear terms and the use of multiple initial beam positions to form a smooth envelope are sensible methodological choices. The paper also clearly addresses an important gap, namely the lack of interpretable machine-learning models for nonlinear topological photonics. However, the present version provides only qualitative comparisons (Figs. 3-5), reports no numerical error measures or coefficient values, and does not test the completeness of the candidate-function library. These omissions leave the central claim about 'proven' freedom from a priori limitations unsupported. The approach is nevertheless promising and worthy of publication after substantial strengthening.

major comments (4)
  1. [Abstract, Section IV, Eq. (4)] The claim that the scheme is 'proven free of the a priori limitations imposed by the underlying hierarchy of scales' overreaches. The library used in Eq. (4) is assembled to fit that specific equation, i.e., a local, low-order (derivatives up to third order) and polynomial (up to quintic) ansatz. This is itself a model assumption: nonlocal couplings, higher derivatives, or coupling to additional transverse degrees of freedom would be forced into the selected terms. Either provide a completeness/robustness test (for example, inject known terms outside the library into a synthetic dataset and show that sparse regression identifies or rejects them correctly) or replace 'proven free' with a more limited claim that is valid only within the chosen function library.
  2. [Sections III and IV, Figs. 3-5] No quantitative validation is reported. The 20% test split is mentioned in Section II, but no test error is given. Figure 3 is a visual comparison between the dispersion obtained from the fitted coefficients and the band structure of the same microscopic lattice model, and Figs. 4 and 5 show representative snapshots without normalized errors. Please report the recovered coefficient values with their uncertainties from the 100 validation folds, the training and test residuals, and a quantitative comparison between the full paraxial simulations and the PDE solutions (e.g., relative L2 error as a function of propagation distance for several beam widths and input powers). Such numbers are necessary to support the statements that the reconstructed equations 'accurately reproduce' the dynamics and that the coefficients 'confirm the presence' of the nonlinear effects.
  3. [Section II and Fig. 3] The validation is in large part circular. The coefficients are fit to paraxial simulation data, and Fig. 3 compares the dispersion obtained from those coefficients with the band structure of the same lattice model; both curves derive from the same underlying model. This is a consistency check rather than an independent prediction. To demonstrate predictive power, hold out complete configurations (for example, one domain-wall shape or one parameter set) or train on one set of beam widths and predict another, and report errors on those held-out cases.
  4. [Section IV, Eq. (5)] The two-step procedure assumes that the linear operator is already correctly determined and then fits only the nonlinear corrections against Eq. (5). If the linear coefficients carry errors, those errors will propagate into the nonlinear coefficients, but the manuscript does not quantify this effect. Please report the values and uncertainties of all nonlinear coefficients (G1, G2, vg1, vg2, vg3, gamma1, gamma2), and either quantify the error propagation or perform a joint fit of the linear and nonlinear parameters.
minor comments (5)
  1. [Section II, text near Fig. 2] The panel references are inconsistent: the text says 'zigzag (c,e) and bearded (b,d)' but the caption and figure layout indicate the lower row should be labeled (d) and (f); please correct the references.
  2. [Throughout the manuscript] There are numerous missing spaces between words, including in the abstract ('Theuseof', 'bandstructures', 'ofinterest'); the manuscript should be carefully proofread.
  3. [Section II, Eq. (2)] Equation (2) uses a tilde for the slowly varying amplitude, whereas the data-extracted envelope is denoted A(z,x) without a tilde; please define the exact relation between the two and use consistent notation throughout.
  4. [Section II] The intensity threshold used to select data points is never specified. Please state the threshold values used for each dataset and, ideally, show that the reconstructed coefficients are insensitive to reasonable variations of this threshold.
  5. [Fig. 3(c)] The text mentions that the approximation is recovered near K+ on one side and near 2π on the other side of the brown vertical line segment, but it is not explained why the model changes branch or what the brown line segment represents; please clarify in the caption or main text.

Circularity Check

1 steps flagged · score 6.0 of 10

Partial circularity: the nonlinear 'effects confirmed' are pre-built into the regression library, so Eq. (5) is the fit ansatz restated and the reconstructed PDE is a parameter fit, not an independent discovery.

  1. fitted input called prediction [Section IV, Eqs. (4)-(5) and following paragraph]
    "The library of functions is assembled to fit the following equation ... − i(vg1 A + vg2|A|^2 A + vg3|A|^4 A) ∂|A|^2/∂x − (G1|A|^2 + G2|A|^4) A ... The determined coefficients confirm the presence of two competing major nonlinear effects. ... This self-steepening deformation occurs due to the prevailing nonlinear velocity term vg [16, 23]."

    The candidate library is deliberately assembled from the same nonlinear terms that appear in the target equation (4): self-focusing terms G1|A|^2+G2|A|^4 and self-steepening terms vg1, vg2, vg3 multiplying ∂|A|^2/∂x. Equation (5) is just the real/imaginary decomposition of those pre-selected terms. Sparse regression then returns nonzero coefficients for these ansatz functions, and the paper describes this as 'confirm[ing] the presence' of self-focusing and self-steepening. The confirmation is circular: the effects were inserted as regression inputs, so nonzero fitted coefficients only show that the chosen ansatz can absorb the data, not that the data independently establish those effects.

full rationale

The paper is transparent that the regression coefficients are fit to simulation data and validated on a held-out 20% test split, so the linear reconstruction and the comparison in Fig. 3 are honest in-sample validations, not hidden fabrications. The main circularity is narrower and specific to the nonlinear regime: the library is 'assembled to fit' Eq. (4), which already contains the self-focusing and self-steepening terms, and Eq. (5) is the same ansatz rewritten. Declaring that the fitted coefficients 'confirm the presence' of these effects is therefore equivalent to saying the pre-chosen terms have nonzero fitted values. This does not invalidate the coefficient-estimation exercise, but it does weaken the claim that the scheme 'determines the structure of the equation' or is free of a priori modeling choices: the functional form is imposed by the library. The 'proven free of the a priori limitations imposed by the underlying hierarchy of scales' statement is an overreach because the finite polynomial derivative library is itself an a priori ansatz, though that is more a correctness concern than a circularity. Self-citations to Refs. [16] and [23] motivate the expected nonlinear terms but are not load-bearing for the fitted coefficients, so they do not independently raise the score. Overall, the central nonlinear 'confirmation' reduces by construction to the fitted ansatz, giving partial circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central model is a fit of many coefficients to simulation data. The most important unstated input is the completeness of the function library and the smoothness of the interpolated envelope, both of which determine whether the fitted PDE is a faithful reduced description.

free parameters (5)
  • linear PDE coefficients (beta0, v, eta, eta') = not reported
    Fit via sparse regression to low-intensity paraxial envelope data; values and uncertainties are not given in the text.
  • nonlinear phase coefficients (G1, G2) = not reported
    Fit to the difference between full evolution and linear operator at high intensity, Eq. (5a).
  • nonlinear velocity coefficients (vg1, vg2, vg3) = not reported
    Fit to the intensity equation, Eq. (5b), capturing self-steepening.
  • gain and loss coefficients (gamma1, gamma2) = not reported
    Fit to the intensity equation, Eq. (5b); no independent evidence these are physical rather than a way to absorb model error.
  • intensity threshold for data selection = not specified
    Chosen by hand to exclude low-intensity points; affects derivative estimates and fitted coefficients.
assumptions (6)
  • domain assumption The paraxial wave equation, Eq. (1), with Kerr-type nonlinearity accurately models the waveguide lattice system.
    Used as the ground truth data generator; standard in optics, but not independently verified in this paper.
  • standard math Numerical solutions from plane wave expansion and beam propagation methods are sufficiently accurate to serve as ground truth.
    No convergence tests are reported, so the accuracy of the reference simulations is assumed.
  • domain assumption The envelope extracted at discrete waveguide centers and interpolated by sweeping initial positions yields a smooth function A(z,x) governed by a local PDE.
    Section II describes this construction; if the envelope is not smooth or the dynamics are nonlocal, the finite-difference derivative library fails.
  • ad hoc to paper The library of candidate functions is complete and local, with derivatives up to third order and polynomial nonlinearities up to quintic.
    Section IV states that the library is assembled to fit Eq. (4), which is based on the prior asymptotic model rather than discovered from data.
  • domain assumption Nonlinear coefficients are independent of z and x over the propagation distances considered.
    Assumed in Eq. (4); no test for z or x dependence is provided.
  • standard math Sparse regression with an 80/20 train-test split identifies the correct coefficients when the library contains the true terms and data noise is low.
    This is the standard SINDy assumption, and the paper does not test behavior under realistic noise levels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-driven model reconstruction for nonlinear wave dynamics." pith.science (2026). https://pith.science/paper/UBWPNYOT

@misc{pith2026241111556,
  author       = {Pith},
  title        = {Pith review of: Data-driven model reconstruction for nonlinear wave dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UBWPNYOT}},
  note         = {Machine review of arXiv:2411.11556}
}
read the original abstract

The use of machine learning to predict wave dynamics is a topic of growing interest, but commonly-used deep learning approaches suffer from a lack of interpretability of the trained models. Here we present an interpretable machine learning framework for analyzing the nonlinear evolution dynamics of optical wavepackets in complex wave media. We use sparse regression to reduce microscopic discrete lattice models to simpler effective continuum models which can accurately describe the dynamics of the wavepacket envelope. We apply our approach to valley-Hall domain walls in honeycomb photonic lattices of laser-written waveguides with Kerr-type nonlinearity and different boundary shapes. The reconstructed equations accurately reproduce the linear dispersion and nonlinear effects including self-steepening and self-focusing. This scheme is proven free of the a priori limitations imposed by the underlying hierarchy of scales traditionally employed in asymptotic analytical methods. It represents a powerful interpretable machine learning technique of interest for advancing design capabilities in photonics and framing the complex interaction-driven dynamics in various topological materials.

Figures

Figures reproduced from arXiv: 2411.11556 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic of the methodology for extracting the PDE model with ML using regression and numerically [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. (a,b) Two distinct shapes of valley-Hall domain [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4. Evolution of the beam envelope at zigzag (left [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: FIG. 3. Comparison between the numerical (blue dots) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]
Figure 5
Figure 5. Figure 5: FIG. 5. The nonlinear scenarios of beam propagation at [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    Machine learning and the physical sciences,

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zde- borov´ a, “Machine learning and the physical sciences,” Rev. Mod. Phys.91, 045002 (2019)

    work page 2019

  2. [2]

    Physics-informed learn- ing of governing equations from scarce data,

    Z. Chen, Y. Liu, and H. Sun, “Physics-informed learn- ing of governing equations from scarce data,” Nature 6 Communications 12, 6136 (2021)

    work page 2021

  3. [3]

    Machine learning inverse problem for topological photonics,

    L. Pilozzi, F. A. Farrelly, G. Marcucci, and C. Conti, “Machine learning inverse problem for topological photonics,” Communications Physics1, 57 (2018)

    work page 2018

  4. [4]

    Deep learning for topological photonics,

    J. Yun, S. Kim, S. So, M. Kim, and J. Rho, “Deep learning for topological photonics,” Advances in Physics: X7, 2046156 (2022)

    work page 2022

  5. [5]

    PySINDy: A com- prehensive Python package for robust sparse system identification,

    A. A. Kaptanoglu, B. M. de Silva, U. Fasel, K. Ka- heman, A. J. Goldschmidt, J. Callaham, C. B. De- lahunt, Z. G. Nicolaou, K. Champion, J.-C. Loiseau, J. N. Kutz, and S. L. Brunton, “PySINDy: A com- prehensive Python package for robust sparse system identification,” Journal of Open Source Software 7, 3994 (2022)

    work page 2022

  6. [6]

    PySINDy: A Python pack- age for the sparse identification of nonlinear dynami- cal systems from data,

    B. de Silva, K. Champion, M. Quade, J.-C. Loiseau, J. Kutz, and S. Brunton, “PySINDy: A Python pack- age for the sparse identification of nonlinear dynami- cal systems from data,” Journal of Open Source Soft- ware 5, 2104 (2020)

    work page 2020

  7. [7]

    Data-driven discovery of partial differential equations,

    S. H. Rudy, S. L. Brunton, J. L. Proctor, and J. N. Kutz, “Data-driven discovery of partial differential equations,” Science Advances3, e1602614 (2017)

    work page 2017

  8. [8]

    SINDy-Pi: a robust algorithm for parallel implicit sparse iden- tification of nonlinear dynamics,

    K.J.N.KahemanKadierdanandB.S.L., “SINDy-Pi: a robust algorithm for parallel implicit sparse iden- tification of nonlinear dynamics,” Journal of Open Source Software 476, 20200279 (2020)

    work page 2020

Show all 23 references
  1. [9]

    Discoveringequationsthat govern experimental materials stability under envi- ronmental stress using scientific machine learning,

    R. R. Naik, A. Tiihonen, J. Thapa, C. Batali, Z. Liu, S.Sun, andT.Buonassisi, “Discoveringequationsthat govern experimental materials stability under envi- ronmental stress using scientific machine learning,” npj Computational Materials8, 72 (2022)

    2022

  2. [10]

    Data-driven model discovery of ideal four-wave mixing in nonlinear fibre optics,

    A. V. Ermolaev, A. Sheveleva, G. Genty, C. Finot, and J. M. Dudley, “Data-driven model discovery of ideal four-wave mixing in nonlinear fibre optics,” Sci- entific Reports 12, 12711 (2022)

    2022

  3. [11]

    Topological photonics,

    T. Ozawa, H. M. Price, A. Amo, N. Goldman, M.Hafezi, L.Lu, M.C.Rechtsman, D.Schuster, J.Si- mon, O. Zilberberg, and I. Carusotto, “Topological photonics,” Rev. Mod. Phys.91, 015006 (2019)

    2019

  4. [12]

    Modulational instability and solitary waves in polariton topological insulators,

    Y. V. Kartashov and D. V. Skryabin, “Modulational instability and solitary waves in polariton topological insulators,” Optica 3, 1228–1236 (2016)

    2016

  5. [13]

    Instability of bosonic topological edge states in the presence of interactions,

    Y. Lumer, M. C. Rechtsman, Y. Plotnik, and M. Segev, “Instability of bosonic topological edge states in the presence of interactions,” Phys. Rev. A 94, 021801 (2016)

    2016

  6. [14]

    Topological edge states and gap solitons in the nonlinear Dirac model,

    D. A. Smirnova, L. A. Smirnov, D. Leykam, and Y. S. Kivshar, “Topological edge states and gap solitons in the nonlinear Dirac model,” Laser & Photonics Re- views 13, 1900223 (2019)

    2019

  7. [15]

    Valley Hall edge solitons in a photonic graphene,

    Q. Tang, B. Ren, V. O. Kompanets, Y. V. Kartashov, Y. Li, and Y. Zhang, “Valley Hall edge solitons in a photonic graphene,” Optics Express29, 39755–39765 (2021)

    2021

  8. [16]

    Gradient catastrophe of nonlinear photonic valley-Hall edge pulses,

    D. A. Smirnova, L. A. Smirnov, E. O. Smolina, D. G. Angelakis, and D. Leykam, “Gradient catastrophe of nonlinear photonic valley-Hall edge pulses,” Physical Review Research 3, 043027 (2021)

    2021

  9. [17]

    Discov- ering governing equations from data by sparse identi- fication of nonlinear dynamical systems,

    S. L. Brunton, J. L. Proctor, and J. N. Kutz, “Discov- ering governing equations from data by sparse identi- fication of nonlinear dynamical systems,” Proceedings of the National Academy of Sciences113, 3932–3937 (2016)

    2016

  10. [18]

    Identification of moment equations via data-driven approaches in nonlinear Schr¨ odinger models,

    S. Yang, S. Chen, W. Zhu, and P. G. Kevrekidis, “Identification of moment equations via data-driven approaches in nonlinear Schr¨ odinger models,” Fron- tiers in Photonics5 (2024)

    2024

  11. [19]

    Observation of photonic topological valley Hall edge states,

    J. Noh, S. Huang, K. P. Chen, and M. C. Rechtsman, “Observation of photonic topological valley Hall edge states,” Phys. Rev. Lett.120, 063902 (2018)

    2018

  12. [20]

    Regression shrinkage and selection via the lasso,

    R. Tibshirani, “Regression shrinkage and selection via the lasso,” Journal of the Royal Statistical Society Series B: Statistical Methodology58, 267–288 (1996)

    1996

  13. [21]

    R. E. Schapire, The Boosting Approach to Machine Learning: An Overview (Springer New York, New York, NY, 2003), pp. 149–171

    2003

  14. [22]

    See Supplemental Material for more detailed infor- mation on data preparation and analysis across both linear and nonlinear regimes

  15. [23]

    Self-steepening-induced stabiliza- tion of nonlinear edge waves at photonic valley-Hall interfaces,

    E. O. Smolina, L. A. Smirnov, D. Leykam, and D. A. Smirnova, “Self-steepening-induced stabiliza- tion of nonlinear edge waves at photonic valley-Hall interfaces,” Phys. Rev. A108, L061501 (2023)

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.