Pith. sign in

REVIEW 4 major objections 5 minor 49 references

SPIN-ODE: Stiff Physics-Informed Neural ODE for Chemical Reaction Rate Estimation

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read SPIN-ODE recovers chemical reaction rate coefficients from stiff trajectories without close initial guesses, and validates this on synthetic and atmospheric benchmarks.

desk verdict A real three-stage training trick for stiff chemical kinetics with strong Robertson/AOXID results, undercut by the POLLU exclusion of the five fastest coefficients and by single-run reporting. read the letter →

arxiv 2505.05625 v3 pith:XAUYHUVY submitted 2025-05-08 cs.LG cs.AI

Wenqing Peng , Zhi-Song Liu , Michael Boy This is my paper · ORCID
classification cs.LGcs.AI
keywords chemicalreactionrateestimationstiffordinarydifferentialequationsneuralphysics-informednetworksnetworkatmosphericchemistrytrajectoryfittingcoefficientdiscovery
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

SPIN-ODE tries to solve an inverse problem: given concentration measurements of a stiff chemical system and a fixed list of reactions, recover the rate coefficients of those reactions without being handed a good starting guess. The paper's central claim is that this can be done by splitting the problem into three stages: first fit a black-box neural ODE to the trajectories, then train a physics-constrained chemical-reaction neural network on interpolated finite-difference derivatives, and finally fine-tune the coefficients by integrating the network as an ODE. On the classic three-reaction stiff benchmark, the 25-reaction POLLU air-pollution scheme, and a 49-reaction autoxidation scheme, it reports log-scale mean absolute errors of $7.4\times 10^{-3}$, $0.46$, and $0.10$, respectively. If correct, this matters because rate coefficients in atmospheric chemistry span many orders of magnitude and are usually fitted from expensive chamber or flow-tube experiments with close initial guesses; an automatic estimator would turn ordinary trajectory data into mechanistic parameters.

What carries the argument

The load-bearing object is the staged training loop rather than any single network. Its two pillars are the normalization layers of the black-box neural ODE, which write $\mathrm{d}y/\mathrm{d}t = NN_\theta((y-y_{\min})/(y_{\max}-y_{\min}), t)\,(y_{\max}-y_{\min})/t_{\text{scale}}$ so that stiffness does not produce runaway gradients, and the CRNN's log-linear parameterization of each rate law, which makes every reaction rate a product of a learnable positive coefficient and the reactant concentrations raised to their stoichiometric powers. The CRNN converts derivative labels into physically interpretable coefficients through the fixed stoichiometric matrix, and the final ODE-integration fine-tuning supplies trajectory-level constraints that derivative matching alone misses.

What would settle it

Generate trajectories from a three-species stiff system with a known true coefficient set, add a slow fourth reaction that the model is not told about, and run the pipeline with the original three-reaction network; if the estimated coefficients for the known reactions stay within the paper's reported error scale the pathway assumption is not load-bearing for those coefficients, but if they shift far beyond that scale the central claim fails for any setting where the pathway set is incomplete.

Watch

Extended reading notes

Core claim

The paper establishes that stiffness can be overcome for rate-coefficient extraction by decoupling trajectory fitting from parameter fitting. In the first stage, a normalized black-box neural ODE learns the concentration trajectories; the normalization maps species concentrations and their time derivatives to comparable ranges, and two derivative-based loss terms stabilize training. The trained trajectory is interpolated to a finer time grid and finite differences turn it into supervised $(\hat{y}, \dot{\hat{y}})$ pairs. A Chemical Reaction Neural Network (CRNN), which encodes each reaction rate as $r_i = \exp(\ln\theta_i + \sum_j s^f_{i,j}\ln[Y_j])$ and forms $\mathrm{d}Y/\mathrm{d}t = S \cdot R$ through the fixed stoichiometric matrix $S$, is then trained on these pairs without any ODE integration. Only after this pre-training is the CRNN re-coupled to a stiff ODE solver and fine-tuned on the full trajectories. The paper reports that this three-stage schedule, especially the interpolation between stages one and two, is what makes the coefficient estimates accurate, and it frames the full pipeline as the first to recover stiff-reaction rate coefficients without close initial guesses.

Load-bearing premise

The method assumes the reaction network, namely the list of reactions and the stoichiometric matrix $S$, is known and correct, and it further assumes the stage-one trajectory fit is accurate enough that finite differences of its interpolated values preserve the rate information; neither assumption is tested under pathway error.

Editorial extensions

If this is right

  • If the central claim holds, measured concentration time series from stiff atmospheric systems can be converted into reaction rate coefficients without a human-provided close initial guess, removing a major obstacle to using machine learning for detailed chemistry.
  • Because the CRNN has only one parameter per reaction, the estimated coefficients are directly readable as rate constants and can be inserted into existing numerical chemical solvers for prediction or mechanism analysis.
  • The reported downsampling results suggest the method degrades gracefully with data sparsity: a tenfold reduction in observations raises the AOXID coefficient error to roughly $0.45$ on the log scale, leaving the estimates physically plausible.
  • The step-two-to-step-three improvement indicates that trajectory-level integration adds information not present in local derivative matching, so the final fine-tuning stage is a necessary part of the pipeline rather than a minor adjustment.

Reading between the lines

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

  • The reported accuracy is conditional on the reaction network being correct; a natural test the paper does not run is to feed a stoichiometric matrix with one missing or spurious reaction and check whether the remaining rate coefficients shift, which would quantify how much of the success comes from the known-pathway prior.
  • The interpolation step's strong effect suggests that the key to stiff inverse problems is obtaining reliable derivative labels rather than solver stability alone; one could test whether higher-order finite differences, smoothing splines, or a learned denoiser on the stage-one trajectory further lowers the coefficient error on sparse data.
  • The POLLU experiment excludes five hydroxyl-cycling coefficients because their values were supplied during training; an editorially interesting extension is to test full 25-coefficient recovery and determine whether the exclusion reflects an identifiability limit of the data or a limitation of the optimization.
  • Because the stage-one MLP is a black box, its approximation error is inherited by the derivative labels; comparing against a stage-one model that enforces mass conservation would separate trajectory-fitting error from rate-recovery error.
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 paper proposes SPIN-ODE, a three-stage neural-ODE framework for estimating chemical reaction rate coefficients from concentration trajectories. Stage 1 fits a black-box (MLP) neural ODE to observed trajectories, using scaled states and derivative-based regularizers. Stage 2 resamples and interpolates the fitted trajectories, computes finite-difference derivative labels, and pre-trains a Chemical Reaction Neural Network (CRNN) to regress these derivatives. Stage 3 couples the pre-trained CRNN to a stiff ODE solver and fine-tunes the rate coefficients against the observed trajectories. The method assumes a fixed, known reaction network (the stoichiometric matrix and the list of reactions). The authors evaluate on the Robertson problem, the POLLU atmospheric chemistry benchmark (25 reactions), and a newly proposed AOXID autoxidation scheme (49 reactions), reporting low trajectory-mean-squared errors and low log-scale mean absolute errors for the estimated rate coefficients, and they provide code.

Significance. If the claims hold, SPIN-ODE is a practically useful contribution: it combines the robustness of a black-box stiff neural ODE with the interpretability of a CRNN, and it appears to recover rate coefficients from stiff synthetic trajectories without close initial guesses. The Robertson and AOXID results are strong, and the code is publicly available (GitHub and archived Zenodo), which is a clear strength. The method's fixed-pathway assumption is an acknowledged limitation, though it is not presented consistently. The central concerns are: (i) the POLLU evaluation excludes five coefficients that were provided during training, which directly weakens the 'absence of prior knowledge' claim; (ii) the pre-training derivative labels are not validated against true derivatives, a load-bearing step for the recovery of fast rate coefficients; and (iii) all results are single runs, so the reported state-of-the-art numbers lack statistical support. These issues are fixable within the manuscript's scope, and the core idea remains defensible.

major comments (4)
  1. [Section 4.2 and Supplementary Table 1] The reported POLLU rate-coefficient error (MAE = 0.46 in Table 1) excludes 5 of the 25 coefficients (reactions 15-19), which the text states were 'provided during training to facilitate convergence.' This is incompatible with the abstract's claim of estimating coefficients 'without prior knowledge.' Moreover, among the remaining 20 coefficients, per-coefficient errors in Supplementary Table 1 are large: reaction 25 (N2O5 = NO3 + NO2) has ground truth 3.12E+00 but prediction 7.1E-02 (a factor of about 44), and reaction 21 is off by a factor of about 3.5. The MAE therefore hides substantial individual failures. Please report per-coefficient errors for all 25 reactions, explain why reactions 15-19 are not estimable from the provided trajectories, and adjust the 'absence of prior knowledge' claim to match what is actually demonstrated.
  2. [Section 3.2, Eq. (15), and Section 4.3] The CRNN pre-training uses derivative labels obtained by finite differences of the interpolated MLP trajectory fitted in Step 1. These labels are not independent measurements of the true kinetics; they inherit any bias in the Step-1 fit. The ablation in Table 2 compares alternative label-generation strategies but never checks the labels against the ground-truth derivatives, which are available because all data are synthetically generated. This matters most for fast reactions: the paper's own POLLU experiment excludes the fastest coefficients (15-19, up to 1e12), which is exactly the regime where slow trajectory observations do not constrain fast rates. I ask the authors to add a direct validation: compare the finite-difference derivatives from the fitted MLP with the analytic derivatives of the known ODE, and report whether the fast transient information is preserved. Without this, the claim of recovering stiff fast coefficients without close initial guesses is not fully supported.
  3. [Abstract, Section 1 contribution bullet, Section 3.2] The paper repeatedly claims estimation 'without prior knowledge' (Abstract, Section 5) and describes the method as 'the first work to automatically find optimal rate coefficients in the absence of prior knowledge.' However, the method requires the full reaction network—the list of reactions and the stoichiometric matrix S—as input (Eqs. 13-14). The contribution bullet itself says 'given reaction pathways,' which is an internal inconsistency in the central claim. The authors should either revise the claim to 'rate coefficients given known pathways' or provide an experiment in which the pathway set is misspecified or inferred, to substantiate the stronger wording.
  4. [Table 1 and Section 4.2] All reported results are single runs; no seed variance or error bars are given. The paper claims 'robustness' (Section 4.4) and state-of-the-art performance, but with stochastic training (Adam, random initialization) a single run is insufficient to establish that the reported MAEs (especially POLLU's 0.46) are representative. Please report mean and standard deviation over at least 5 seeds for the main tables, or justify why the results are deterministic. This is a reproducibility requirement for a journal submission.
minor comments (5)
  1. [Section 4.2, Figure 3 caption] The text 'POLL Y' should be 'POLLU'.
  2. [Equation (12)] The weights α and β are fixed at 0.1 without sensitivity analysis. A brief statement or a small ablation on their values would strengthen the claim that they are robust.
  3. [Section 4.1] The sliding window size (20) and stride (10) for POLLU and AOXID are chosen without ablation; the robustness study in Table 3 varies data sparsity but not these hyperparameters.
  4. [Section 4.2 and Reference [22]] The comparison with Ji et al. [22] would be fairer if the authors noted that Ji et al. targets autonomous pathway discovery in addition to rate estimation; the current text implies a like-for-like comparison.
  5. [Section 4.4, Table 3] A 10x downsampling increases the rate-coefficient error from 0.145 to 0.437; the word 'moderate' is subjective. Reporting per-coefficient errors in this setting would help the reader assess whether the degradation is uniform or concentrated in a few fast reactions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the rate-coefficient estimates are trajectory fits, not re-used inputs.

full rationale

The derivation is not circular. For Robertson and AOXID, the target rate coefficients are never used as inputs: the datasets are generated by numerical integration of the mass-action ODEs (Eqs. 3-5), and the SPIN-ODE pipeline (Eqs. 10-15 plus the step-3 trajectory integration loss) recovers coefficients by fitting concentrations, so the final estimates are constrained by the observed trajectories rather than by the ground-truth constants. The CRNN architecture encodes the known stoichiometric matrix and the mass-action rate law, which is an explicit modeling assumption rather than a hidden reintroduction of the target coefficients; the claim of operating 'in the absence of prior knowledge' is overstated because reaction pathways are provided, but that is an overclaim, not a circular reduction. Two evaluation limitations are disclosed in the paper: (i) POLLU coefficients 15-19, the stiff OH-cycling subset, were 'provided during training to facilitate convergence' and excluded from the reported error (Section 4.2), so the POLLU result is not a fully prior-knowledge-free test; and (ii) the Section 3.2 Step-2 derivative labels are finite differences of the Step-1 MLP's own interpolated trajectory (Eq. 15), so CRNN pretraining is supervised by a learned surrogate rather than by independent derivative measurements. Neither limitation makes the final s3 coefficient estimates reduce to the inputs by construction: Step 3 integrates the CRNN against the trajectory loss, and the provided POLLU coefficients are not counted as predictions. There is no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation; the paper is self-contained as an inverse problem on simulation-generated benchmarks.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method's central output is obtained by minimizing a trajectory loss over CRNN parameters that are the rate coefficients themselves. Nothing is derived analytically; the load is carried by the assumed network structure, the known stoichiometry, the numerical solvers, and the hand-set loss weights and window parameters. The POLLU benchmark also receives five true coefficients during training, which are then excluded from the reported error.

free parameters (3)
  • Loss weights alpha and beta = 0.1, 0.1
    Set in Eq. 12 to balance trajectory, velocity, and acceleration losses; no sensitivity analysis is reported.
  • Sliding window size and stride = 20, 10
    Chosen for POLLU and AOXID to reduce integration error; only the downsampling robustness study in Section 4.4 varies observation density, not window parameters.
  • POLLU OH cycling coefficients (reactions 15-19) = Ground truth from literature
    Provided during training and excluded from the estimation result, making them inputs rather than fitted outputs; this weakens the no-prior-knowledge claim.
assumptions (4)
  • domain assumption Reaction pathways and stoichiometric coefficients are known and fixed.
    The CRNN in Eqs. 13-14 requires a fixed stoichiometric matrix S; all experiments assume correct pathways, and no experiment tests wrong or incomplete pathways.
  • domain assumption Chemical kinetics follows power-law mass-action rate laws with constant rate coefficients.
    The CRNN architecture encodes Eq. 3, and the synthetic datasets are generated from the same law, so the validation is within the same model class.
  • standard math The numerical stiff solvers (BDF, Rosenbrock, Kvaerno3) produce accurate reference solutions and gradients.
    Used to generate ground-truth trajectories and to integrate the CRNN in step 3; solver error is assumed negligible.
  • ad hoc to paper Finite differences on the interpolated fitted trajectory approximate the true derivatives well enough for CRNN pre-training.
    Step 2 uses Eq. 15; the ablation shows interpolation helps, but no error bound or theoretical guarantee is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SPIN-ODE: Stiff Physics-Informed Neural ODE for Chemical Reaction Rate Estimation." pith.science (2026). https://pith.science/paper/XAUYHUVY

@misc{pith2026250505625,
  author       = {Pith},
  title        = {Pith review of: SPIN-ODE: Stiff Physics-Informed Neural ODE for Chemical Reaction Rate Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XAUYHUVY}},
  note         = {Machine review of arXiv:2505.05625}
}
read the original abstract

Estimating rate coefficients from complex chemical reactions is essential for advancing detailed chemistry. However, the stiffness inherent in real-world atmospheric chemistry systems poses severe challenges, leading to training instability and poor convergence, which hinder effective rate coefficient estimation using learning-based approaches. To address this, we propose a Stiff Physics-Informed Neural ODE framework (SPIN-ODE) for chemical reaction modelling. Our method introduces a three-stage optimisation process: first, a black-box neural ODE is trained to fit concentration trajectories; second, a Chemical Reaction Neural Network (CRNN) is pre-trained to learn the mapping between concentrations and their time derivatives; and third, the rate coefficients are fine-tuned by integrating with the pre-trained CRNN. Extensive experiments on both synthetic and newly proposed real-world datasets validate the effectiveness and robustness of our approach. As the first work addressing stiff neural ODE for chemical rate coefficient discovery, our study opens promising directions for integrating neural networks with detailed chemistry.

Figures

Figures reproduced from arXiv: 2505.05625 by the authors.

Figure 1
Figure 1. The proposed SPIN-ODE fits concentration trajectories (a) to infer reaction rate coefficients (b), converging from random initialisation to the true data (light colour → dark colour). dependencies and chemical reaction paradigms, into a learnable and generalisable model architecture. Moreover, the scarcity of real experimental data further compli￾cates this endeavour. Most available measurements come from highly con… view at source ↗
Figure 2
Figure 2. Our proposed SPIN-ODE framework for stiff reaction rate coefficient estimation. It includes three steps: 1) black-box neural ODE fitting by optimising predicted ODE trajectories, 2) resampling data from step 1 and pre-training physics-based CRNN by optimising predicted time derivatives, and 3) fine-tuning CRNN by optimising predicted concentration trajectories. The detailed structure of the CRNN is shown in the uppe… view at source ↗
Figure 3
Figure 3. Concentration trajectory fitting compared with observation. From left to right are the results for Robertson, POLLU and AOXID problems. 3 representative concentration trajectories are selected from the POLLU and AOXID problems. Results for Robertson show a gradual approximation to ground truth with our incremental approach [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Estimated rate coefficients for Robertson, POLLU and AOXID, from left to right. The blue cross symbol represents the model prediction, and the yellow circle represents the ground truth value. nents rather than the full system dynamics. This further indicates the necess…
Figure 1
Figure 1. Figure 1: 3 Complete concentration trajectory fitting for POLLU and AOXID problems. The complete concentration trajectory fittings for POLLU (20 chem￾ical reactions) and AOXID (44 chemical reactions) problems are shown in [PITH_FULL_IMAGE:figures/full_fig_p009_1.png]
Figure 1
Figure 1. Figure 1: Convergence of SPIN-ODE on the POLLU dataset: step 1 (trajectory fitting), step 2 (CRNN pre-training), and step 3 (CRNN fine-tuning). Loss is shown on a logarithmic scale [PITH_FULL_IMAGE:figures/full_fig_p010_1.png]
Figure 2
Figure 2. Figure 2: Concentration trajectory fitting for the POLLU problem [PITH_FULL_IMAGE:figures/full_fig_p012_2.png]
Figure 3
Figure 3. Figure 3: Concentration trajectory fitting for the AOXID problem [PITH_FULL_IMAGE:figures/full_fig_p013_3.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 33 canonical work pages

  1. [1]

    Betancourt, C

    C. Betancourt, C. W. Y. Li, F. Kleinert, and M. G. Schultz. Graph machine learning for improved imputation of missing tropospheric ozone data. Environmental Science & Technology, 57 0 (46): 0 18246--18258, 2023. doi:10.1021/acs.est.3c05104. PMID: 37661931

    work page doi:10.1021/acs.est.3c05104 2023

  2. [2]

    K. Bi, L. Xie, H. Zhang, and et al. Accurate medium-range global weather forecasting with 3d neural networks. Nature, 619: 0 533--538, 2023

    work page 2023

  3. [3]

    Bodnar, W

    C. Bodnar, W. P. Bruinsma, and et al. Aurora: A foundation model of the atmosphere. arXiv, 2405.13063, 2024

    arXiv 2024

  4. [4]

    J. Butcher. A history of runge-kutta methods. Applied Numerical Mathematics, 20 0 (3): 0 247--260, 1996. ISSN 0168-9274. doi:https://doi.org/10.1016/0168-9274(95)00108-5

    doi:10.1016/0168-9274(95)00108-5 1996

  5. [5]

    T. Q. Chen, Y. Rubanova, J. Bettencourt, and D. K. Duvenaud. Neural ordinary differential equations. Advances in neural information processing systems, 2018

    work page 2018

  6. [6]

    B. Chu, T. Chen, and et al. Application of smog chambers in atmospheric process studies. National Science Review, 9 0 (2): 0 nwab103, 06 2021. ISSN 2095-5138. doi:10.1093/nsr/nwab103

    work page doi:10.1093/nsr/nwab103 2021

  7. [7]

    Cuomo, V

    S. Cuomo, V. Di Cola, and F. e. a. Giampaolo. Scientific machine learning through physics–informed neural networks: Where we are and what’s next. J Sci Comput, 92 0 (88), 2022

    work page 2022

  8. [8]

    Curtiss and J

    C. Curtiss and J. Hirschfelder. Integration of stiff equations. Proceedings of the National Academy of Sciences of the United States of America, 38 0 (3), 1952

    work page 1952

Show all 49 references
  1. [9]

    Damian, A

    V. Damian, A. Sandu, M. Damian, F. Potra, and G. R. Carmichael. The kinetic preprocessor KPP -a software environment for solving chemical kinetics. Computers & Chemical Engineering, 26 0 (11): 0 1567--1579, Nov. 2002. ISSN 00981354. doi:10.1016/S0098-1354(02)00128-X

    2002 doi

  2. [10]

    Q. Di, H. Amini, and et al. Assessing no2 concentration and model uncertainty with high spatiotemporal resolution across the contiguous united states using ensemble model averaging. Environmental Science & Technology, 54 0 (3): 0 1372--1384, 2020. doi:10.1021/acs.est.9b03358. ...

    2020 doi

  3. [11]

    Dupont, A

    E. Dupont, A. Doucet, and Y. W. Teh. Augmented neural odes. Advances in neural information processing systems, 2019

    2019

  4. [12]

    Finlay, J.-H

    C. Finlay, J.-H. Jacobsen, L. Nurbekyan, and A. Oberman. How to train your neural ode: the world of jacobian and kinetic regularization. In International conference on machine learning, pages 3154--3164. PMLR, 2020

    2020

  5. [13]

    Fronk and L

    C. Fronk and L. Petzold. Training stiff neural ordinary differential equations with implicit single-step methods. Chaos: An Interdisciplinary Journal of Nonlinear Science, 34 0 (12): 0 123147, 12 2024

    2024

  6. [14]

    Fronk and L

    C. Fronk and L. Petzold. Training stiff neural ordinary differential equations with explicit exponential integration methods. Chaos: An Interdisciplinary Journal of Nonlinear Science, 35 0 (3): 0 033154, 03 2025

    2025

  7. [15]

    Ghosh, H

    A. Ghosh, H. Behl, E. Dupont, P. Torr, and V. Namboodiri. Steer: Simple temporal regularization for neural odes. Advances in neural information processing systems, 2020

    2020

  8. [16]

    Goswami, A

    S. Goswami, A. D. Jagtap, H. Babaee, B. T. Susi, and G. E. Karniadakis. Learning stiff chemical kinetics using extended deep neural operators. Computer Methods in Applied Mechanics and Engineering, 419: 0 116674, 2024

    2024

  9. [17]

    L. Hou, Q. Dai, C. Song, and et al. Revealing drivers of haze pollution by explainable machine learning. Environmental Science & Technology Letters, 9 0 (2): 0 112--119, 2022. doi:10.1021/acs.estlett.1c00865

    2022 doi

  10. [18]

    C. J. Howard. Kinetic measurements using flow tubes. The Journal of Physical Chemistry, 83: 0 3--9, 1979

    1979

  11. [19]

    Huang, K

    X. Huang, K. Kandris, and E. Katsou. Training stiff neural ordinary differential equations in data-driven wastewater process modelling. Journal of Environmental Management, 373: 0 123870, Jan. 2025. ISSN 03014797. doi:10.1016/j.jenvman.2024.123870

    2025

  12. [20]

    S. I. and B. T. Speeding up astrochemical reaction networks with autoencoders and neural odes. Advances in neural information processing systems Workshop, 2023

    2023

  13. [21]

    Ingebrand, A

    T. Ingebrand, A. Thorpe, and U. Topcu. Zero-shot transfer of neural odes. Advances in Neural Information Processing Systems, 37: 0 67604--67626, 2024

    2024

  14. [22]

    Ji and S

    W. Ji and S. Deng. Autonomous discovery of unknown reaction pathways from data by chemical reaction neural network. The Journal of Physical Chemistry A, 125 0 (4): 0 1082--1092, Feb. 2021. ISSN 1089-5639, 1520-5215. doi:10.1021/acs.jpca.0c09316

    2021 doi

  15. [23]

    W. Ji, W. Qiu, Z. Shi, S. Pan, and S. Deng. Stiff-pinn: Physics-informed neural network for stiff chemical kinetics. The Journal of Physical Chemistry A, 125 0 (36): 0 8098--8106, Sept. 2021. ISSN 1089-5639, 1520-5215. doi:10.1021/acs.jpca.1c05102

    2021 doi

  16. [24]

    P. Kidger. On neural differential equations. arXiv preprint arXiv:2202.02435, 2022

    2022 arXiv

  17. [25]

    S. Kim, W. Ji, S. Deng, Y. Ma, and C. Rackauckas. Stiff neural ordinary differential equations. Chaos: An Interdisciplinary Journal of Nonlinear Science, 31 0 (9): 0 093122, Sept. 2021. ISSN 1054-1500, 1089-7682. doi:10.1063/5.0060697

    2021 doi

  18. [26]

    B. C. Koenig, S. Kim, and S. Deng. Kan-odes: Kolmogorov–arnold network ordinary differential equations for learning dynamical systems and hidden physics. Computer Methods in Applied Mechanics and Engineering, 432: 0 117397, 2024. ISSN 0045-7825. doi:https://doi.org/10.1016/j.c...

    2024

  19. [27]

    B. C. Koenig, S. Kim, and S. Deng. Chemkans for combustion chemistry modeling and acceleration. arXiv preprint arXiv:2504.12580, 2025

    2025 arXiv

  20. [28]

    X. Kong, K. Yamashita, B. Foggo, and N. Yu. Dynamic parameter estimation with physics-based neural ordinary differential equations. In 2022 IEEE Power & Energy Society General Meeting ( PESGM ) , pages 1--5, Denver, CO, USA, July 2022. IEEE. ISBN 978-1-6654-0823-3. doi:10.1109...

    2022

  21. [29]

    Lam and et al

    R. Lam and et al. Learning skillful medium-range global weather forecasting. Science, 382: 0 1416--1421, 2023

    2023

  22. [30]

    X. Li, J. Zhang, Q. Zhu, C. Zhao, X. Zhang, X. Duan, and W. Lin. From fourier to neural odes: Flow matching for modeling complex systems. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pages 2939...

    2024

  23. [31]

    Z. Liu, Y. Wang, S. Vaidya, F. Ruehle, J. Halverson, M. Solja c i \'c , T. Y. Hou, and M. Tegmark. Kan: Kolmogorov-arnold networks. arXiv preprint arXiv:2404.19756, 2024

    2024 arXiv

  24. [32]

    Z.-S. Liu, P. Clusius, and M. Boy. Neural network emulator for atmospheric chemical ode. Neural Networks, 184: 0 107106, 2025. ISSN 0893-6080. doi:https://doi.org/10.1016/j.neunet.2024.107106

    2025

  25. [33]

    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 0 (3): 0 218--229, 2021

    2021

  26. [34]

    Massaroli, M

    S. Massaroli, M. Poli, J. Park, A. Yamashita, and H. Asama. Dissecting neural odes. Advances in neural information processing systems, 2020

    2020

  27. [35]

    Owoyele and P

    O. Owoyele and P. Pal. Chemnode: A neural ordinary differential equations framework for efficient chemical kinetic solvers. Energy and AI, 7: 0 100118, 2022. ISSN 2666-5468. doi:https://doi.org/10.1016/j.egyai.2021.100118

    2022

  28. [36]

    W. Peng. Spin-ode: Stiff physics-informed neural ode for chemical reaction rate estimation (v1.0.0), 2025. URL https://doi.org/10.5281/zenodo.16949166

    2025 doi

  29. [37]

    Peng, Z.-S

    W. Peng, Z.-S. Liu, and M. Boy. Spin-ode: Stiff physics-informed neural ode for chemical reaction rate estimation. arXiv preprint arXiv:2505.05625, 2025. Full version of this paper

    2025 arXiv

  30. [38]

    Pichelstorfer, S

    L. Pichelstorfer, S. P. O'Meara, and G. McFiggans. A theory-informed, experiment-based constraint on the rate of autoxidation chemistry--an analytical approach. Aerosol Research, 3 0 (2): 0 417--428, 2025

    2025

  31. [39]

    Robertson and J

    H. Robertson and J. Walsh. Numerical analysis: an introduction. In Chapitre the Solution of a Set of Reaction Rate Equations. Academic Press, 1966

    1966

  32. [40]

    Roesch, C

    E. Roesch, C. Rackauckas, and M. P. H. Stumpf. Collocation based training of neural ordinary differential equations. Statistical Applications in Genetics and Molecular Biology, 20 0 (2): 0 37--49, July 2021. ISSN 1544-6115, 2194-6302. doi:10.1515/sagmb-2020-0025

    2021 doi

  33. [41]

    Salunkhe, D

    A. Salunkhe, D. Deighan, P. E. DesJardin, and V. Chandola. Chemtab: A physics guided chemistry modeling framework. In International Conference on Computational Science, pages 75--88. Springer, 2022

    2022

  34. [42]

    Seiler, W

    E. Seiler, W. Lei, and P. Protopapas. Stiff transfer learning for physics-informed neural networks. arXiv preprint arXiv:2501.17281, 2025

    2025 arXiv

  35. [43]

    X. Su, W. Ji, J. An, Z. Ren, S. Deng, and C. K. Law. Kinetics parameter optimization of hydrocarbon fuels via neural ordinary differential equations. Combustion and Flame, 251: 0 112732, May 2023. ISSN 00102180. doi:10.1016/j.combustflame.2023.112732

    2023

  36. [44]

    A. C. Th\"oni, W. E. Robinson, Y. Bachrach, W. T. Huck, and T. Kachman. Modeling chemical reaction networks using neural ordinary differential equations. Journal of Chemical Information and Modeling, 2025

    2025

  37. [45]

    J. G. Verwer. Gauss–seidel iteration for stiff odes from chemical kinetics. SIAM Journal on Scientific Computing, 15 0 (5): 0 1243--1250, Sept. 1994. ISSN 1064-8275, 1095-7197. doi:10.1137/0915076

    1994 doi

  38. [46]

    Yildiz, M

    C. Yildiz, M. Heinonen, and H. Lahdesmaki. Ode2vae: Deep generative second order odes with bayesian neural networks. Advances in Neural Information Processing Systems, 32, 2019

    2019

  39. [47]

    D. D. Yulia Rubanova, Ricky T. Q. Chen. Latent odes for irregularly-sampled time series. Advances in neural information processing systems, 2019

    2019

  40. [48]

    Q. Zhu, J. L. Laughner, and R. C. Cohen. Combining machine learning and satellite observations to predict spatial and temporal variation of near surface oh in north american cities. Environmental Science & Technology, 56 0 (11): 0 7362--7371, 2022. doi:10.1021/acs.est.1c05636....

    2022 doi

  41. [49]

    B. Zou, E. Levine, P. Zaharieva, R. Johari, and E. Fox. Hybrid ^2 neural ODE causal modeling and an application to glycemic response. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pages 62934--6...

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.