Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Long-term simulation of physical and mechanical behaviors using curriculum-transfer-learning based physics-informed neural networks

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Splitting a long simulation into short windows whose predictions feed the next window keeps physics-informed neural networks accurate over long horizons.

desk verdict A credible, incremental extension of time-sliced PINN training, with honest limitations and a useful real-data demo—worth a review, not a breakthrough. read the letter →

arxiv 2502.07325 v1 pith:ZNXRBVGG submitted 2025-02-11 cs.LG cs.NAmath.NA

Yuan Guo , Zhuojia Fu , Jian Min , Shiyu Lin , Xiaoting Liu , Youssef F. Rashed , Xiaoying Zhuang This is my paper · ORCID
classification cs.LGcs.NAmath.NA
keywords physics-informedneuralnetworkscurriculumlearningtransferlong-termsimulationtime-domaindecompositioninverseproblemsThreeGorgesReservoirAreaSaint-Venantequations
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

Standard physics-informed neural networks (PINNs) lose accuracy and stability when asked to solve time-dependent partial differential equations over long horizons. This paper proposes CTL-PINN, a training schedule that decomposes a long simulation into short subproblems, solves the first with a standard PINN, then advances the solution by alternating two strategies: curriculum learning, which extends the trained time window in small increments, and transfer learning, which stacks new time windows onto the learned model. The claim is that this combination yields more accurate and stable long-term solutions than standard PINN, and does so in fewer steps than using either strategy alone. The method is demonstrated on nonlinear wave propagation, the dynamic response of a Kirchhoff plate, and a hydrodynamic model of the Three Gorges Reservoir Area, including inverse estimation of source terms and roughness coefficients.

What carries the argument

The central object is CTL-PINN, a closed training loop whose state is the trained network from the previous time window. What carries the argument is the additional supervised learning dataset $\tau_{sp}^{(k)}$: each new subproblem is trained not only on the PDE residual, initial, and boundary data, but also on the previous network's predictions at points in the overlap of the old and new time domains (Eqs. (13) and (18), with losses (16) and (22)). Curriculum learning uses these points to widen the window in small steps; transfer learning reuses parameter initialization plus predicted initial conditions to stack non-overlapping windows. The practical rule that curriculum steps stay bounded by $T_p/4$ and transfer steps by $T_{ct}/4$ is what keeps the recycled predictions close enough to the true solution to serve as teaching signals.

What would settle it

On a benchmark with a known analytical solution, such as the nonlinear annulus wave problem with $T=250$ s, run CTL-PINN and record the $L^2$ error after every curriculum and transfer step. If the per-step error grows superlinearly---for instance, if doubling the number of steps more than doubles the accumulated error---then the recycled predictions are not accurate enough to sustain the claimed long-horizon gain.

Watch

Extended reading notes

Core claim

The paper's central claim is that the failure of standard PINNs on long-horizon evolution problems can be overcome by a two-phase training schedule that reuses the previous model as both initialization and teacher. In the curriculum phase, the network trained on $(0,T_p]$ is fine-tuned on successively larger windows $(0,T_p+k\Delta T_{pc}]$, with predictions of the previous network on the overlap region added as extra supervised points (Eqs. (13) and (16)). In the transfer phase, the model from $(0,T_c]$ is carried forward window by window, using predicted values at the new window's initial time as its initial condition and overlap predictions as teaching data (Eqs. (18), (19), and (22)). The resulting method is reported to keep $L^2$ errors small where standard PINN diverges, to reach a given horizon with fewer training steps than curriculum-only or transfer-only variants, and to extend the same machinery to inverse problems and to field data with limited measurements.

Load-bearing premise

The load-bearing premise is that the previous network's predictions, recycled as additional supervised learning points and as initial conditions for the next window, are accurate enough to guide the next subproblem; if their errors are too large, the errors compound and the claimed advantage over standard PINN disappears, as the paper itself acknowledges when it notes that there is a certain error between the predicted data and the exact values.

Editorial extensions

If this is right

  • For a fixed long horizon, CTL-PINN reaches a stable solution with fewer training steps than curriculum-only (CL-PINN) or transfer-only (TL-PINN) variants, because curriculum learning establishes a larger accurate base window before transfer stacking.
  • The same schedule handles second-order time derivatives and systems of PDEs, including inverse problems: source-term inversion for nonlinear waves and roughness-coefficient inversion for the reservoir both stay accurate where standard PINN drifts.
  • Adding real measurements at intermediate times reduces the error accumulation caused by recycled predictions, so the method is directly applicable to data-scarce field problems.
  • For new problems, the paper's recommended bounds---curriculum step at most $T_p/4$ and transfer step at most $T_{ct}/4$---give a concrete starting configuration.
  • In the Three Gorges case, the method produces 24-hour-ahead predictions of discharge and water level from hydrological-station data while calibrating the roughness parameter.

Reading between the lines

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

  • The fixed step-size recommendations ($\Delta T_{pc}\le T_p/4$ and $\Delta T_{ct}\le T_{ct}/4$) are heuristics; a natural extension the paper does not pursue is adaptive step selection driven by the measured mismatch between recycled predictions and any available observations.
  • If the recycled predictions stay accurate, the same scheme could serve as an online data-assimilation loop for operational river forecasting: each window injects sparse gauge data and corrects drift before the next extrapolation step.
  • The plate example suggests the schedule also alleviates time-causality violations in continuous PINNs, but the paper does not quantify this; comparing CTL-PINN against causality-weighted loss formulations on the same benchmarks would isolate the mechanism.
  • A stress test beyond the reported 24-hour forecast horizon, or with rainfall-induced lateral inflow where the model currently sets $q_l=U_l=0$, would show whether the short-term constraint effect that powers extrapolation persists under stronger nonlinearity.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CTL-PINN, a curriculum-transfer-learning variant of physics-informed neural networks for long-time simulation of time-dependent PDEs. The method decomposes the time domain into short windows, trains a standard PINN on the first window, then repeatedly extends the domain by using the previous network's predictions as additional supervised data (Eqs. (13), (18)) and as initial conditions for the next transfer window (Eq. (19)). The authors claim that this combination overcomes the instability and poor accuracy of standard PINNs over long horizons, requires fewer steps than curriculum-only or transfer-only variants, and extends to inverse problems. The method is demonstrated on nonlinear wave propagation, Kirchhoff plate dynamics, and a Saint-Venant hydrodynamic model of the Three Gorges Reservoir Area, including roughness-coefficient inversion and 24-hour-ahead prediction.

Significance. If the central claim holds, CTL-PINN is a useful and simple extension of existing time-domain-decomposition PINN methods, and the Three Gorges application is a noteworthy real-world demonstration of physics-informed learning with limited measured data. The paper is honest about the main weakness: Conclusion (2) explicitly acknowledges that self-generated supervised points carry error and lead to error accumulation. No error bound or convergence analysis is provided, the numerical comparisons are based on single runs with hand-tuned step sizes, and no code is released. The method is therefore best viewed as an empirically promising training schedule rather than a proven algorithmic improvement. The three examples, especially the reservoir case with independent measured data, give qualitative support to the claim, but the evidence base is not yet strong enough for the paper's broad conclusions.

major comments (4)
  1. [Section 3.1-3.2, Eqs. (13), (18)-(19), Conclusion (2)] The distinguishing mechanism of CTL-PINN is the reuse of the previous model's predictions as additional supervised data and as transfer-window initial conditions. The paper itself acknowledges in Conclusion (2) that these pseudo-labels contain error and cause error accumulation, but it provides no bound, contraction estimate, or numerical control showing that the pseudo-label error stays small relative to the residual and data losses over the reported horizons. Consequently, the claimed advantage over standard PINN could in principle be an artifact of favorable step sizes rather than a property of the method. I request per-step reporting of the pseudo-label error against the reference solution, an ablation where pseudo-labels are replaced by exact or independent reference data for the first several windows, or a formal accumulation estimate.
  2. [Section 4, Table 3, Figs. 5, 8, 12] All reported L2 errors appear to come from a single training run for each configuration. The methods use stochastic optimizers (Adam in Section 4.3, L-BFGS initialized randomly in Sections 4.1 and 4.2) and randomly sampled Latin Hypercube points, so the claim that CTL-PINN is 'stable' and 'robust' is not supported without repeated-seed statistics. Please report means and standard deviations over at least several independent runs, or, if the runs are intended to be deterministic, state the seeding and initialization procedure explicitly and justify why the comparison is not run-to-run dependent.
  3. [Algorithm 1, Sections 3.1-3.2] The method's performance depends on the free hyperparameters ΔT_pc, ΔT_ct, the curriculum step count n, and the transfer step count m. The text gives only heuristic guidance (ΔT_pc ≤ T_p/4, T_p < T_c ≤ 3T_p, ΔT_ct ≤ T_c/4) and one configuration per example. No sensitivity analysis is provided, so it is unclear how robust the reported advantage over curriculum-only and transfer-only baselines is to these choices. Please include a systematic variation of step sizes and step counts around the chosen values, and show how the final L2 errors and step counts change.
  4. [Section 4.3.1, Eqs. (34)-(35)] Section 4.3.1 states that the reservoir model 'does not have initial and boundary conditions,' yet the Saint-Venant system is an initial-boundary-value problem and the subsequent text describes discharge data at segment endpoints and water level at a hydrological station. The mathematical role of these time series is unclear: are endpoint discharges used as boundary conditions, is the hydrological-station water level only an observation, and how is the state initialized at t=0? Please specify exactly which terms enter the loss function (Eqs. (41)-(44)) and how the problem is regularized, since this directly affects the reproducibility and validity of the inversion and 24-hour prediction claims.
minor comments (6)
  1. [Abstract, Section 4.1.1, Fig. 5] The naming is inconsistent: the abstract refers to 'CL-PINN' for curriculum learning only, while Section 4.1.1 and Figure 5 use 'CT-PINN' for the same variant; please unify the terminology.
  2. [Section 4.1, Eq. (27)] The displayed source term Q(x,y,z,t) is garbled by missing operators and parentheses, making it difficult to verify that the analytical solution satisfies the governing equation; please re-typeset the equation.
  3. [Section 3.2, Eq. (19)] The initial-condition dataset formula in Eq. (19) is badly formatted and partially unreadable; please rewrite it so that the time at which the source-model predictions are evaluated is clear.
  4. [Section 4.2, Figs. 13-14] The text refers to an 'analytical solution' for the Kirchhoff plate, but the captions and text also refer to the 'ABAQUS solution'; please clarify whether ABAQUS is the reference and, if so, state the mesh and solver accuracy used to generate it.
  5. [Section 4.3, Table 3] The columns 'Error Q' and 'Error Z' are not defined; please specify whether these are relative L2 errors, absolute errors, or another metric, and give the test-set time interval used for the error computation.
  6. [Data availability] The data availability statement says code for sections other than 4.3 is available 'upon request'; for a computational paper, releasing code and data in a repository would substantially improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are validated against independent analytical, ABAQUS, and held-out measured references.

full rationale

The paper's central claims (CTL-PINN improves accuracy and stability over standard PINN for long-time simulation) are not derived from the method's own outputs. In each example, the reported L2 errors (Eq. 23) are computed against independent reference values: the analytical solution in Section 4.1, the ABAQUS solution in Section 4.2, and held-out measured water-level/discharge data in Section 4.3. The training loop does use the previous network's predictions as additional supervised learning points (Eqs. 13-16 and 18-22), which is self-referential in mechanism, and the authors explicitly acknowledge that this causes error accumulation in Conclusion (2). However, this is a stated limitation and a potential failure mode, not a case where the claimed prediction reduces by construction to an input: the final models are still constrained by PDE residuals, initial/boundary conditions, and measured data, and the evaluation benchmarks are external. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling citation is used. The comparisons with CL-PINN and TL-PINN are empirical, not definitional. Therefore, the derivation chain is self-contained with respect to the paper's main conclusions.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method's central free parameters are curriculum and transfer step sizes and counts, chosen by heuristics. The key self-referential assumption is using the model's own previous output as training data. The physical application additionally assumes sparse surveyed sections represent the reach and that lateral inflow can be neglected. No new physical entities, forces, particles, or dimensions are introduced.

free parameters (4)
  • Curriculum learning step size ΔT_pc = 5 s (Sections 4.1, 4.2), 1 h (Section 4.3)
    Recommended by the heuristic ΔT_pc ≤ T_p/4; not derived. Directly controls error accumulation.
  • Transfer learning step size ΔT_ct = 10 s (Sections 4.1, 4.2), 2 h (Section 4.3)
    Recommended by the heuristic ΔT_ct ≤ T_c/4; not derived from any error bound.
  • Curriculum and transfer step counts (n, m) = (6, 20), (4, 6), (12, 6)
    Chosen per problem to keep reported errors small; no selection rule or sensitivity study is given.
  • Roughness coefficient n in Section 4.3 = 0.0435 (Cuntan-Tongluoxia), 0.0316 (Changshou-Weidong)
    Inverted from measured data as an optimizable parameter. It is a target of the method, not an ad hoc physical assumption.
assumptions (5)
  • ad hoc to paper The previous-step PINN prediction u_theta^{(k-1)} used as additional supervised data is sufficiently accurate to guide the next time window.
    Equations (13) and (18) build the supervised dataset from the model's own predictions; the paper admits this causes error accumulation.
  • domain assumption Time-domain decomposition with overlapping intervals preserves the PDE solution and causality.
    Sections 3.1 and 3.2 assume parameter import and overlap supervision across subdomains are valid without a proof of continuity or consistency.
  • domain assumption The cross-section interpolation functions f_B, f_A, f_R from three or four measured sections represent the whole river reach.
    Appendix equations (47) to (53) average three or four sections and apply them to the entire segment; sparse sections may not capture the actual geometry.
  • domain assumption Ignoring lateral inflow q_l and rainfall is acceptable for the test period.
    Section 4.3.2 states there are no major tributaries and no significant rainfall; this is a data-period assumption, not a universal property.
  • domain assumption PINN optimization with L-BFGS or Adam reaches a sufficiently low training loss.
    The whole comparison presumes the networks are trained to convergence; no convergence certificates or random seeds are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Long-term simulation of physical and mechanical behaviors using curriculum-transfer-learning based physics-informed neural networks." pith.science (2026). https://pith.science/paper/ZNXRBVGG

@misc{pith2026250207325,
  author       = {Pith},
  title        = {Pith review of: Long-term simulation of physical and mechanical behaviors using curriculum-transfer-learning based physics-informed neural networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZNXRBVGG}},
  note         = {Machine review of arXiv:2502.07325}
}
read the original abstract

This paper proposes a Curriculum-Transfer-Learning based physics-informed neural network (CTL-PINN) for long-term simulation of physical and mechanical behaviors. The main innovation of CTL-PINN lies in decomposing long-term problems into a sequence of short-term subproblems. Initially, the standard PINN is employed to solve the first sub-problem. As the simulation progresses, subsequent time-domain problems are addressed using a curriculum learning approach that integrates information from previous steps. Furthermore, transfer learning techniques are incorporated, allowing the model to effectively utilize prior training data and solve sequential time domain transfer problems. CTL-PINN combines the strengths of curriculum learning and transfer learning, overcoming the limitations of standard PINNs, such as local optimization issues, and addressing the inaccuracies over extended time domains encountered in CL-PINN and the low computational efficiency of TL-PINN. The efficacy and robustness of CTL-PINN are demonstrated through applications to nonlinear wave propagation, Kirchhoff plate dynamic response, and the hydrodynamic model of the Three Gorges Reservoir Area, showcasing its superior capability in addressing long-term computational challenges.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Physical Informed Neural Networks for modeling ocean pollutant

    cs.LG 2025-07 conditional novelty 4.0 of 10

    A physics-informed neural network trained in Julia fits noisy finite-difference data for a constant-velocity 2D advection-diffusion problem and reports an 8.25% relative L2 error against the same solver's clean output.

Reference graph

Works this paper leans on

5 extracted references · 4 canonical work pages · cited by 1 Pith paper

  1. [1]

    Arf, J., Reichle, M., Klinkel, S., & Simeon, B. (2023). Scaled boundary isogeometric analysis with C1 coupling for Kirchhoff plate theory. Computer Met ods in Applied Mec nics nd Engineering, 415, 116198. Beck, C., Becker, S., Grohs, P., Jaafari, N., & Jentzen, A. (2021). Solving the Kolmogorov PDE by means of deep learning. Journ l of Scientific Computin...

    work page 2023

  2. [9]

    Xiao, S., Bai, J., Jeong, H., Alzubaidi, L., & Gu, . (2025). A meshless Runge -Kutta-based Physics-Informed Neural Network framework for structural vibration analysis. Engineering An l sis wit Bound r Elements, 17 , 106054. i, ., Tang, C., ang, Z., Zhang, S., & Zhang, C. (2017). A One‐Dimensional Hydrodynamic and Water Quality Model for a Water Transfer P...

    work page 2025

  3. [64]

    Islam, A., Raghuwanshi , N., & Singh, R. (2008). Development and application of hydraulic simulation model for irrigation canal network. Journ l of Irrig tion nd Dr in ge Engineering, 134(1), 49–59. Jagtap, A. D., & Karniadakis, G. E. (2020). Extended physics-informed neural networks (XPINNs): A generalized space- time domain decomposition based deep lear...

    work page Pith review arXiv 2008

  4. [1003]

    C., Taylor, R

    Zienkiewicz, O. C., Taylor, R. L., Nithiarasu, P., & Zhu, J. (1977). T e finite element met od (V ol. 3). Elsevier

    work page 1977

  5. [5237]

    Venutelli, M. (2002). Stability and accuracy of weighted four -point implicit finite difference schemes for open channel flow. Journ l of H dr ulic Engineering, 128(3), 281–288. 31 Wang, S., Teng, ., & Perdikaris, P. (2020). Underst nding nd mitig ting gr dient p t ologies in p sics -informed neur l networks (arXiv:2001.04536). arXiv. Weiss, K., Khoshgoft...

    arXiv 2002

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.