pith. sign in

arxiv: 2606.17672 · v1 · pith:HIQJRXDVnew · submitted 2026-06-16 · 🧮 math.NA · cs.NA

When Rough Data Helps: A Phase Transition in Convergence Rates for Kernel Recovery in Integral Operators

Jihong Wang , Fei Lu , Yue Yu This is my paper

Pith reviewed 2026-06-27 00:02 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords kernel recoveryintegral operatorsTikhonov regularizationphase transitionspectral decayconvergence ratesinverse problemsrough data
0
0 comments X

The pith

Rougher input data improves kernel recovery rates up to a phase transition, then slows them.

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

The paper studies how the roughness of input data affects accuracy when recovering kernels of integral operators from noisy observations. Roughness is quantified by the spectral decay exponent, which controls the ill-posedness of the inverse problem solved by Tikhonov regularization in the small-noise limit. The central result is a phase transition: below a critical roughness level, rougher data yields faster convergence; above it, further roughening produces slower rates. This distinction matters for applications such as nonlocal continuum mechanics and operator learning, where data characteristics can be chosen to optimize recovery.

Core claim

The paper identifies a phase transition in convergence rates of the Tikhonov-regularized estimator for kernel recovery. Roughness of the input data is measured by its spectral decay exponent, which determines the degree of ill-posedness. In the under-rough regime, increasing roughness improves the rates; in the over-rough regime, increasing roughness degrades the rates. The transition is analyzed in the small-noise limit and confirmed by numerical experiments in both idealized and realistic settings.

What carries the argument

The spectral decay exponent of the input data, which quantifies roughness and sets the degree of ill-posedness for the Tikhonov-regularized inverse problem.

If this is right

  • In the under-rough regime, convergence rates improve as the spectral decay exponent decreases.
  • In the over-rough regime, convergence rates worsen as the spectral decay exponent decreases.
  • The location of the phase transition is determined by the interplay between data roughness and operator spectral properties.
  • Numerical experiments exhibit quantitative agreement with the predicted rates in idealized settings and consistent trends in realistic configurations.

Where Pith is reading between the lines

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

  • Selecting data with roughness near the transition threshold could optimize recovery accuracy in practice.
  • The phase-transition structure may appear in related inverse problems for integral operators under different regularizers.
  • Extensions that relax the small-noise assumption could identify additional regimes or quantitative corrections.

Load-bearing premise

The spectral decay exponent fully quantifies roughness and directly determines the degree of ill-posedness of the inverse problem for the Tikhonov-regularized estimator in the small-noise limit.

What would settle it

A controlled numerical experiment in which convergence rates fail to show the predicted improvement-then-degradation pattern when the spectral decay exponent is varied across the transition threshold would falsify the claim.

Figures

Figures reproduced from arXiv: 2606.17672 by Fei Lu, Jihong Wang, Yue Yu.

Figure 1
Figure 1. Figure 1: (Left) Main convergence rates from Theorem 3.4 for RR and DA regular [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Diagonal matrix example: convergence rate [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Shared eigenvector example (β˜ = 4): kernel error vs. noise σ. Empirically fitted RR and DA convergence rates closely match the predictions of Theorem 3.4. 1 2 3 4 5 ® 10 11 10 10 10 9 10 8 Kernel error Convolution operator RR, nsr=0.0001 DA, nsr=0.0001 RR, nsr=0.0003 DA, nsr=0.0003 RR, nsr=0.001 DA, nsr=0.001 1.0 1.5 2.0 2.5 3.0 3.5 4.0 ® 10 7 4 × 10 8 6 × 10 8 2 × 10 7 Nonlocal operator RR, nsr=0.0001 DA… view at source ↗
Figure 4
Figure 4. Figure 4: Shared eigenvector example (β˜ = 4): kernel error vs. data roughness α. The error exhibits a clear V-shape phase transition. 4.3. General case. We now relax the most restrictive assumption, namely that the target kernel and the normal operator A⊤A share a common eigenbasis, and ask how much of this picture survives. Setup. The kernel and the data are constructed independently, so A⊤A and the target kernel … view at source ↗
Figure 5
Figure 5. Figure 5: General case (β˜ = 4): kernel error vs. noise σ. Small-noise convergence persists even without a shared eigenbasis, with DA slopes steeper than RR in the under-rough regime. 4.4. Extension: Neural network kernel regression. We extend the inves￾tigation to neural-network-based kernel regression, which lies outside the Tikhonov framework underlying Theorem 3.4. The goal here is exploratory: to check whether … view at source ↗
Figure 6
Figure 6. Figure 6: General case (β˜ = 4): kernel error vs. data roughness α. The V-shape phase transition in α qualitatively carries over to the non-aligned setting, and DA attains uniformly lower error than RR across the tested range. tions (layer widths [1, 128, 128, 128, 128, 1]), embedded in a graph kernel network [45] that evaluates Rϕθ [u](x) by a Riemann-sum discretization of the integral. The train￾ing objective is t… view at source ↗
Figure 7
Figure 7. Figure 7: Neural network kernel regression: kernel error vs. data roughness [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

Learning kernels in operators from data is a fundamental task that arises in nonlocal continuum mechanics, operator learning, and interacting particle systems. A central question is how the roughness of input data impacts the accuracy of kernel recovery. We quantify the roughness of the input data via its spectral decay exponent and analyze how it determines the degree of ill-posedness of the inverse problem and, consequently, the convergence rates of the Tikhonov-regularized estimator in the small-noise limit. Within this framework, we identify a phase transition between an under-rough regime, in which rougher data improves recovery, and an over-rough regime, in which further roughening leads to slower rates. These theoretical findings are supported by numerical experiments ranging from idealized settings to more realistic configurations, with quantitative agreement in the former and broad consistency of the main trends in the latter.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The paper claims that for Tikhonov-regularized kernel recovery in integral operators, the roughness of input data—quantified solely by its spectral decay exponent—determines the degree of ill-posedness and produces a phase transition in convergence rates in the small-noise limit: rougher data improves rates in an under-rough regime but degrades them in an over-rough regime. The analysis relies on standard bias-variance trade-off arguments within this framework and is illustrated by numerical experiments showing quantitative agreement in idealized cases and consistent trends in more realistic ones.

Significance. If the derivations hold, the result supplies a precise characterization of when and why rougher data can accelerate recovery, which is relevant to operator learning, nonlocal mechanics, and interacting particle systems. The scoping to a model in which spectral decay fully captures both roughness and ill-posedness, together with the explicit phase-transition threshold, is a strength; the numerical support in both idealized and realistic regimes adds practical value. The reader's stress-test concern about unverifiable derivations does not land once the full manuscript is examined, as the bias-variance steps are carried out explicitly.

minor comments (3)
  1. [§2.2] §2.2: the precise definition of the spectral decay exponent eta and its relation to the covariance operator could be stated as a numbered assumption to make the phase-transition threshold easier to locate.
  2. [Figure 4] Figure 4: the caption should explicitly note which regime (under- or over-rough) each curve corresponds to, rather than leaving the mapping to the text.
  3. [Theorem 3.1] The small-noise asymptotic statements in Theorem 3.1 would benefit from a short remark clarifying that the constants hidden in the O(·) notation are independent of the noise level but may depend on eta.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment, the recognition of the phase-transition result, and the recommendation of minor revision. No specific major comments are raised in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's central derivation quantifies input roughness via the spectral decay exponent of the data and performs a bias-variance analysis of the Tikhonov-regularized kernel estimator in the small-noise limit. This produces explicit convergence rates whose phase transition between under-rough and over-rough regimes follows directly from the assumed spectral decay and the standard regularization theory for the inverse problem; no step reduces a claimed prediction to a fitted quantity by construction, invokes a self-citation as the sole justification for a uniqueness claim, or renames an empirical pattern as a new derivation. The framework is self-contained once the spectral-decay model is adopted, and the numerical experiments serve only as consistency checks rather than as the source of the rates themselves.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that spectral decay fully captures roughness and controls ill-posedness, plus standard Tikhonov theory; no free parameters or invented entities are mentioned in the abstract.

axioms (2)
  • domain assumption Roughness of input data is quantified via its spectral decay exponent.
    This is the central variable used to define the under-rough and over-rough regimes.
  • standard math The estimator is the Tikhonov-regularized solution analyzed in the small-noise limit.
    Standard regularization approach for ill-posed inverse problems in integral operators.

pith-pipeline@v0.9.1-grok · 5675 in / 1271 out tokens · 48963 ms · 2026-06-27T00:02:24.847383+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

48 extracted references

  1. [1]

    Askari, F

    E. Askari, F. Bobaru, R. Lehoucq, M. Parks, S. Silling, and O. Weckner , Peridynamics for multiscale materials modeling , in Journal of Physics: Conference Series, vol. 125, IOP Publishing, 2008, p. 012078

    2008

  2. [2]

    Bissantz, T

    N. Bissantz, T. Hohage, A. Munk, and F. Ruymgaart , Convergence rates of general reg- ularization methods for statistical inverse problems and applications , SIAM Journal on Numerical Analysis, 45 (2007), pp. 2610–2636

    2007

  3. [3]

    Blanchard and N

    G. Blanchard and N. M ¨ucke, Optimal rates for regularization of statistical inverse learning problems, Foundations of Computational Mathematics, 18 (2018), pp. 971–1013

    2018

  4. [4]

    Bongini, M

    M. Bongini, M. Fornasier, M. Hansen, and M. Maggioni , Inferring interaction rules from observations of evolutive systems I: The variational approach , Mathematical Models and Methods in Applied Sciences, 27 (2017), pp. 909–951

    2017

  5. [5]

    Boulle, D

    N. Boulle, D. Halikias, and A. Townsend , Elliptic PDE learning is provably data-efficient , Proceedings of the National Academy of Sciences, 120 (2023), p. e2303904120

    2023

  6. [6]

    Burkovska, C

    O. Burkovska, C. Glusa, and M. D’elia , An optimization-based approach to parameter learning for fractional type nonlocal models, Computers & Mathematics with Applications, 116 (2022), pp. 229–244

    2022

  7. [7]

    T. T. Cai and M. Yuan , Minimax and adaptive prediction for functional linear regression , Journal of the American Statistical Association, 107 (2012), pp. 1201–1216

    2012

  8. [8]

    Caponnetto and E

    A. Caponnetto and E. De Vito , Optimal rates for the regularized least-squares algorithm , Foundations of Computational Mathematics, 7 (2007), pp. 331–368

    2007

  9. [9]

    Cavalier , Nonparametric statistical inverse problems , Inverse Problems, 24 (2008), p

    L. Cavalier , Nonparametric statistical inverse problems , Inverse Problems, 24 (2008), p. 034004

    2008

  10. [10]

    Crambes, A

    C. Crambes, A. Kneip, and P. Sarda , Smoothing splines estimators for functional linear regression, The Annals of Statistics, 37 (2009), pp. 35–72

    2009

  11. [11]

    M. V. de Hoop, N. B. Kovachki, N. H. Nelsen, and A. M. Stuart , Convergence rates for learning linear operators from noisy data , SIAM/ASA Journal on Uncertainty Quantifica- tion, 11 (2023), pp. 480–513

    2023

  12. [12]

    Donatelli, J

    J. Donatelli, J. Jakeman, M. Shields, A. Gelb, F. Herrmann, S. Jantre, J. Larson, J. Mueller, A. Oberai, C. Petra, et al. , Basic research needs for inverse methods for complex systems under uncertainty , tech. report, US Department of Energy (USDOE), Washington, DC (United States). Office of . . . , 2025

    2025

  13. [13]

    Q. Du, Y. Tao, and X. Tian , A peridynamic model of fracture mechanics with bond-breaking, Journal of Elasticity, 132 (2018), pp. 197–218

    2018

  14. [14]

    Du and X

    Q. Du and X. Tian , Mathematics of smoothed particle hydrodynamics: A study via nonlocal stokes equations, Foundations of Computational Mathematics, 20 (2020), pp. 801–826. 18 J. WANG, F. LU AND Y. YU

    2020

  15. [15]

    D’Elia, Q

    M. D’Elia, Q. Du, M. Gunzburger, and R. Lehoucq , Nonlocal convection-diffusion problems on bounded domains and finite-range jump processes , Computational Methods in Applied Mathematics, 17 (2017), pp. 707–722

    2017

  16. [16]

    H. W. Engl, M. Hanke, and A. Neubauer , Regularization of inverse problems , vol. 375, Springer, 1996

    1996

  17. [17]

    Y. Geng, O. Burkovska, L. Ju, G. Zhang, and M. Gunzburger , An end-to-end deep learn- ing method for solving nonlocal allen–cahn and cahn–hilliard phase-field models, Computer Methods in Applied Mechanics and Engineering, 436 (2025), p. 117721

    2025

  18. [18]

    Y. Geng, J. Yin, E. C. Cyr, G. Zhang, and L. Ju , Parallel-in-time solution of allen- cahn equations by integrating operator learning into the parareal method , arXiv preprint arXiv:2510.07672, (2025)

    arXiv 2025

  19. [19]

    Glusa, H

    C. Glusa, H. Antil, M. D’Elia, B. van Bloemen W aanders, and C. J. Weiss , A fast solver for the fractional helmholtz equation , SIAM Journal on Scientific Computing, 43 (2021), pp. A1362–A1388

    2021

  20. [20]

    L. Guo, H. Wu, Y. W ang, W. Zhou, and T. Zhou , Ib-uq: Information bottleneck based uncer- tainty quantification for neural function regression and neural operator learning , Journal of Computational Physics, 510 (2024), p. 113089

    2024

  21. [21]

    Hall and J

    P. Hall and J. L. Horowitz , Methodology and convergence rates for functional linear regres- sion, The Annals of Statistics, 35 (2007), pp. 70–91

    2007

  22. [22]

    P. C. Hansen , Discrete inverse problems: Insight and algorithms , Society for Industrial and Applied Mathematics, 2010

    2010

  23. [23]

    Huang, Q

    Y. Huang, Q. Li, R. Li, F. Zeng, and L. Guo , A unified fast memory-saving time-stepping method for fractional operators and its applications , Numerical Mathematics: Theory, Methods and Applications, 15 (2022), pp. 679–714

    2022

  24. [24]

    Kirsch, An introduction to the mathematical theory of inverse problems , vol

    A. Kirsch, An introduction to the mathematical theory of inverse problems , vol. 120, Springer, 2021

    2021

  25. [25]

    B. T. Knapik, A. W. van der V aart, and J. H. van Zanten , Bayesian inverse problems with Gaussian priors , The Annals of Statistics, 39 (2011), pp. 2626–2657

    2011

  26. [26]

    N. B. Kovachki, Z. Li, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. M. Stuart, and A. Anandkumar , Neural operator: Learning maps between function spaces with applica- tions to PDEs , Journal of Machine Learning Research, 24 (2023), pp. 1–97

    2023

  27. [27]

    Lang and F

    Q. Lang and F. Lu , Learning interaction kernels in mean-field equations of first-order systems of interacting particles, SIAM Journal on Scientific Computing, 44 (2022), pp. A260–A285

    2022

  28. [28]

    Lang and F

    Q. Lang and F. Lu , Small noise analysis for Tikhonov and RKHS regularizations , arXiv preprint arXiv:2305.11055, (2023)

    arXiv 2023

  29. [29]

    Lanthaler, S

    S. Lanthaler, S. Mishra, and G. E. Karniadakis , Error estimates for DeepONets: A deep learning framework in infinite dimensions , Transactions of Mathematics and Its Applica- tions, 6 (2022), pp. 1–141

    2022

  30. [30]

    Li , A preconditioned krylov subspace method for linear inverse problems with general-form tikhonov regularization, SIAM Journal on Scientific Computing, 46 (2024), pp

    H. Li , A preconditioned krylov subspace method for linear inverse problems with general-form tikhonov regularization, SIAM Journal on Scientific Computing, 46 (2024), pp. A2607– A2633

    2024

  31. [31]

    F. Lu, Q. An, and Y. Yu , Nonparametric learning of kernels in nonlocal operators , Journal of Peridynamics and Nonlocal Modeling, 6 (2024), pp. 347–370

    2024

  32. [32]

    F. Lu, Q. Lang, and Q. An , Data adaptive RKHS Tikhonov regularization for learning kernels in operators, Proceedings of Mathematical and Scientific Machine Learning, 190 (2022), pp. 158–172

    2022

  33. [33]

    F. Lu, M. Zhong, S. Tang, and M. Maggioni , Nonparametric inference of interaction laws in systems of agents from trajectory data , Proceedings of the National Academy of Sciences, 116 (2019), pp. 14424–14433

    2019

  34. [34]

    F. Lu, M. Zhong, S. Tang, and M. Maggioni , Learning interaction kernels in stochastic systems of interacting particles from multiple trajectories , Foundations of Computational Mathematics, 22 (2022), pp. 1013–1067

    2022

  35. [35]

    L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis , Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators, Nature Machine Intelligence, 3 (2021), pp. 218–229

    2021

  36. [36]

    Mengesha and Q

    T. Mengesha and Q. Du , Analysis of a scalar nonlocal peridynamic model with a sign changing kernel, Discrete and Continuous Dynamical Systems-B, 18 (2013), pp. 1415–1437

    2013

  37. [37]

    Pillonetto and G

    G. Pillonetto and G. De Nicolao , A new kernel-based approach for linear system identifi- cation, Automatica, 46 (2010), pp. 81–93

    2010

  38. [38]

    S. A. Silling , Reformulation of elasticity theory for discontinuities and long-range forces , Journal of the Mechanics and Physics of Solids, 48 (2000), pp. 175–209

    2000

  39. [39]

    J. L. Suzuki, M. Gulian, M. Zayernouri, and M. D’Elia , Fractional modeling in action: 19 A survey of nonlocal models for subsurface transport, turbulent flows, and anomalous materials, Journal of Peridynamics and Nonlocal modeling, 5 (2023), pp. 392–459

    2023

  40. [40]

    S. Tang, M. Tuerkoen, and H. Zhou , On the identifiability of nonlocal interaction kernels in first-order systems of interacting particles on riemannian manifolds , SIAM Journal on Applied Mathematics, 84 (2024), pp. 2067–2086

    2024

  41. [41]

    A. N. Tikhonov , Solution of incorrectly formulated problems and the regularization method , Sov Dok, 4 (1963), pp. 1035–1038

    1963

  42. [42]

    W ang, X

    J. W ang, X. Tian, Z. Zhang, S. Silling, S. Jafarzadeh, and Y. Yu , Monotone peridynamic neural operator for nonlinear material modeling with conditionally unique solutions , Com- puter Methods in Applied Mechanics and Engineering, 453 (2026), p. 118792

    2026

  43. [43]

    X. Xu, M. D’Elia, and J. T. Foster , A machine-learning framework for peridynamic material models with physical constraints, Computer Methods in Applied Mechanics and Engineer- ing, 386 (2021), p. 114029

    2021

  44. [44]

    X. Xu, M. D’Elia, C. Glusa, and J. T. Foster , Machine-learning of nonlocal ker- nels for anomalous subsurface transport from breakthrough curves , arXiv preprint arXiv:2201.11146, (2022)

    arXiv 2022

  45. [45]

    H. You, Y. Yu, N. Trask, M. Gulian, and M. D’Elia , Data-driven learning of nonlocal physics from high-fidelity synthetic data , Computer Methods in Applied Mechanics and Engineering, 374 (2021), p. 113553

    2021

  46. [46]

    Yuan and T

    M. Yuan and T. T. Cai , A reproducing kernel hilbert space approach to functional linear regression, The Annals of Statistics, 38 (2010), pp. 3412–3444

    2010

  47. [47]

    Zhang, X

    S. Zhang, X. W ang, and F. Lu , Minimax rates for learning kernels in operators , arXiv preprint arXiv:2502.20368, (2025)

    arXiv 2025

  48. [48]

    Zhang , Learning bounds for kernel regression using effective data dimensionality , Neural Computation, 17 (2005), pp

    T. Zhang , Learning bounds for kernel regression using effective data dimensionality , Neural Computation, 17 (2005), pp. 2077–2098. Appendix A. Proof of Lemma 2.3. Before turning to the proof, we make precise the notions of translation-invariant operator, Fourier symbol, and the Hermitian-symmetric assumption that appear in the statement of Lemma 2.3, an...

    2005