Hybrid Neural Ordinary Differential Equations for Data-Efficient Polymerization Modeling with Incomplete Kinetics

Alexander Mitsos; Eike Cramer; Marah Almanasreh

arxiv: 2606.02145 · v1 · pith:HLGBCPDPnew · submitted 2026-06-01 · 💻 cs.LG

Hybrid Neural Ordinary Differential Equations for Data-Efficient Polymerization Modeling with Incomplete Kinetics

Marah Almanasreh , Alexander Mitsos , Eike Cramer This is my paper

Pith reviewed 2026-06-28 15:59 UTC · model grok-4.3

classification 💻 cs.LG

keywords hybrid neural ODEpolymerization modelingdata-efficient learningincomplete kineticsfree-radical polymerizationmethyl methacrylateneural ordinary differential equations

0 comments

The pith

A hybrid neural ODE keeps explicit physical equations for known polymerization reactions and learns only the radical concentration term to predict from sparse noisy data.

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

The paper aims to show that polymerization dynamics can be modeled accurately by retaining mechanistic mass balances for initiator decomposition, propagation, and termination while using a neural network to learn only the effective radical concentration that drives monomer consumption. This matters for early process design because full mechanistic models require extensive kinetic parameterization and pure data-driven models need large datasets that are expensive to collect. The hybrid is tested on methyl methacrylate batch polymerization with as few as ten measurements under regular and irregular sampling, showing lower errors and better extrapolation than fully data-driven baselines. In a test with noisy data and unseen conditions the hybrid reaches RMSE 0.013 versus 0.31 and 0.68 for the alternatives.

Core claim

The hybrid Neural ODE framework retains explicit mechanistic mass balances for the established reactions while replacing only the partially characterized effective radical concentration with a neural network surrogate. This separation allows the model to achieve reliable generalization from limited data, as evidenced by consistently lower prediction errors and more physically consistent behavior than a pure data-driven NODE or a discrete-time network when both are trained on sparse measurements.

What carries the argument

The hybrid Neural ODE in which the right-hand side is assembled from explicit physical mass-balance equations for known reactions plus a neural network that supplies only the missing radical concentration term.

If this is right

Training on as few as ten measurements yields lower RMSE than either a pure data-driven NODE or a discrete-time network.
The hybrid produces more physically consistent predictions when extrapolating to unseen operating conditions.
Learning only the closure term for radical concentration is sufficient for reliable prediction under noisy and limited data.
Performance remains stable under both regular and irregular sampling schedules.

Where Pith is reading between the lines

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

The same split between known physical terms and a learned surrogate could be tried on other reaction networks where only a subset of rates is well characterized.
If the approach works for batch reactors it could be tested on continuous-flow systems to check whether the learned radical term transfers across reactor types.
The method might shorten the experimental campaign needed for initial process screening by focusing data collection on the unknown term only.

Load-bearing premise

That the known reactions can be modeled accurately with physical equations and that the only missing piece is an effective radical concentration term that a neural network can learn from data.

What would settle it

An experiment in which the hybrid model is trained and tested on data where the assumed physical reaction rates are deliberately misspecified would show whether the performance advantage disappears.

Figures

Figures reproduced from arXiv: 2606.02145 by Alexander Mitsos, Eike Cramer, Marah Almanasreh.

**Figure 2.** Figure 2: Irregular sampling, single-trajectory results. Model predictions are compared to the mechanis [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗

**Figure 3.** Figure 3: Generalization under noisy data. Model predictions for (a) the training trajectory with noisy [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗

read the original abstract

Accurate prediction of polymerization dynamics is essential for process design, control, and optimization. Yet, purely mechanistic models require labor-intensive parameterization of partially characterized kinetics, while purely data-driven models demand large, diverse datasets that are costly to obtain, particularly in early-design stages. We propose a hybrid Neural Ordinary Differential Equation (NODE) framework for data-efficient modeling of free-radical polymerization. Using batch polymerization of methyl methacrylate (MMA) as a case study, the mechanistic mass balances are retained explicitly, and only the partially-characterized effective radical concentration governing monomer consumption is learned from data through a neural network surrogate, while established reactions such as initiator decomposition, propagation, and termination remain physically modeled. The hybrid NODE is evaluated against a discrete-time feedforward neural network and a purely data-driven NODE under sparse data conditions, with models trained on as few as ten measurements under both regular and irregular sampling. The hybrid NODE consistently achieves lower prediction errors and more physically consistent extrapolations than both purely data-driven baselines. In a generalization scenario with noisy data and unseen operating conditions, the hybrid NODE achieves an RMSE of 0.013, compared to 0.31 for the data-driven NODE and 0.68 for the discrete-time model, demonstrating that learning only a closure term rather than the full dynamics is sufficient for reliable prediction under limited data availability.

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.

Desk Editor's Note private letter to a colleague

Hybrid NODE helps on sparse MMA data but missing pure mechanistic baseline weakens the claim that the retained physics is doing the work.

read the letter

The punchline is that this hybrid approach delivers better generalization than pure data-driven models on the reported MMA tests with sparse noisy data, yet the lack of a pure mechanistic baseline leaves open whether the physical equations are actually contributing or just being propped up by the neural term.

What is new here is the concrete construction for polymerization: they retain the mechanistic mass balances for initiator decomposition, propagation, and termination, and learn only a neural surrogate for the effective radical concentration that drives monomer consumption. This targets the incomplete kinetics problem directly instead of replacing the whole model.

The paper does well on the evaluation side. It tests against a discrete-time network and a full data-driven NODE, using as few as ten measurements under regular and irregular sampling. The hybrid reaches an RMSE of 0.013 on the noisy unseen-conditions generalization set, compared to 0.31 and 0.68 for the baselines. That gap is meaningful for the data-efficiency claim.

The soft spot is the missing pure mechanistic run. The strongest claim depends on the retained equations being sufficiently accurate on their own, with the neural part only handling the unknown radical term. Without that baseline evaluated on the same generalization data, any mismatch in the physical parameters could be absorbed by the surrogate, which undercuts the assertion that learning only the closure term is what enables the reliable prediction. The abstract states the mechanistic parts are kept explicit, but the numbers do not separate their contribution.

This is minor in the sense that it is a common evaluation gap rather than a contradiction in the math, but it is central to the paper's argument.

The paper is aimed at chemical engineers modeling polymerization processes with limited data and partial kinetics knowledge. A reader who needs to extrapolate under new operating conditions with few experiments would get concrete value from the comparison.

It deserves a serious referee because the hybrid framing is grounded in domain knowledge and the results are specific enough to discuss. I would send it to peer review.

Referee Report

2 major / 0 minor

Summary. The paper proposes a hybrid Neural ODE framework for free-radical polymerization modeling that retains explicit mechanistic mass balances for initiator decomposition, propagation, and termination while using a neural network surrogate to learn only the partially-characterized effective radical concentration. Using MMA batch polymerization as a case study with as few as ten sparse measurements, the hybrid model is compared to a discrete-time feedforward NN and a purely data-driven NODE, reporting superior performance (RMSE 0.013 vs. 0.31 and 0.68) in a noisy generalization scenario with unseen operating conditions.

Significance. If the central claim holds, the selective hybrid approach could advance data-efficient modeling in chemical engineering by leveraging domain knowledge for most kinetics while learning only a closure term, reducing the data requirements compared to fully data-driven methods and improving physical consistency in extrapolation.

major comments (2)

[Abstract] Abstract: The claim that 'learning only a closure term rather than the full dynamics is sufficient for reliable prediction' is load-bearing for the reported RMSE advantage, yet no results are provided for the pure mechanistic model (identical equations and parameters with the neural surrogate disabled) on the same generalization set with noisy unseen conditions. Without this baseline, it remains possible that the NN compensates for inaccuracies in the retained physical equations rather than solely addressing the radical concentration term.
[Abstract] The evaluation relies on the premise that the mechanistic sub-models for initiator decomposition, propagation, and termination are sufficiently complete and correctly parameterized; however, no sensitivity analysis or validation of these sub-models (e.g., via parameter perturbation or comparison to literature values) is described to confirm they do not require the NN to absorb systematic errors.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and indicate the planned revisions.

read point-by-point responses

Referee: [Abstract] Abstract: The claim that 'learning only a closure term rather than the full dynamics is sufficient for reliable prediction' is load-bearing for the reported RMSE advantage, yet no results are provided for the pure mechanistic model (identical equations and parameters with the neural surrogate disabled) on the same generalization set with noisy unseen conditions. Without this baseline, it remains possible that the NN compensates for inaccuracies in the retained physical equations rather than solely addressing the radical concentration term.

Authors: We agree that the pure mechanistic baseline is necessary to substantiate the central claim. The revised manuscript will include results for the pure mechanistic model (neural surrogate disabled, using the standard quasi-steady-state approximation for radical concentration) evaluated on the identical noisy unseen-conditions test set. This addition will allow direct comparison and clarify the source of the reported performance gains. revision: yes
Referee: [Abstract] The evaluation relies on the premise that the mechanistic sub-models for initiator decomposition, propagation, and termination are sufficiently complete and correctly parameterized; however, no sensitivity analysis or validation of these sub-models (e.g., via parameter perturbation or comparison to literature values) is described to confirm they do not require the NN to absorb systematic errors.

Authors: We acknowledge that explicit sensitivity analysis and parameter validation were omitted from the original submission. The revised version will add a sensitivity study in which the mechanistic rate constants are perturbed within literature-reported uncertainty ranges, with the resulting effect on hybrid-model predictions quantified. We will also explicitly cite the literature sources and nominal parameter values employed for the retained sub-models. revision: yes

Circularity Check

0 steps flagged

No significant circularity; hybrid structure separates independent mechanistic equations from learned closure term.

full rationale

The paper retains explicit mechanistic mass balances for initiator decomposition, propagation, and termination drawn from established domain knowledge, while the neural network surrogate is applied only to the effective radical concentration term. This separation means the reported generalization RMSE on unseen noisy data is not equivalent to the training inputs by construction, nor does any step reduce to a fitted parameter renamed as prediction or a self-citation chain. The central claim rests on the independent physical equations plus data-driven closure, with evaluation on held-out conditions providing external validation rather than tautology. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work are present in the derivation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that only the radical concentration term is unknown and learnable while other kinetics are physically accurate; the neural network parameters are fitted to data but no other free parameters are declared.

free parameters (1)

neural network weights for radical concentration surrogate
Learned from the sparse measurement data to close the mechanistic mass balances.

axioms (1)

domain assumption Mechanistic mass balances for initiator decomposition, propagation, and termination are accurate and can be retained explicitly.
Stated directly in the abstract as the basis for the hybrid structure.

pith-pipeline@v0.9.1-grok · 5771 in / 1317 out tokens · 26145 ms · 2026-06-28T15:59:00.091485+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

53 extracted references · 33 canonical work pages · 1 internal anchor

[1]

Ohshima, M

M. Ohshima, M. Tanigaki, Quality control of polymer production processes, Journal of Process Control 10 (2) (2000) 135–148.doi:https://doi.org/10.1016/S0959-1524(99)00042-6

work page doi:10.1016/s0959-1524(99)00042-6 2000
[2]

Karuppusamy, R

M. Karuppusamy, R. Thirumalaisamy, S. Palanisamy, S. Nagamalai, E. E. S. Massoud, N. Ayrilmis, A review of machine learning applications in polymer composites: advancements, challenges, and future prospects, Journal of Materials Chemistry A 13 (22) (2025) 16290–16308.doi:10.1039/ D5TA00982K

2025
[3]

J. M. Asua, Emulsion polymerization: From fundamental mechanisms to process developments, Journal of Polymer Science Part A: Polymer Chemistry 42 (5) (2004) 1025–1041.doi:https: //doi.org/10.1002/pola.11096

work page doi:10.1002/pola.11096 2004
[5]

Farajzadehahary, S

K. Farajzadehahary, S. Hamzehlou, N. Ballard, Adding machine learning to the polymer reaction engineering toolbox, Progress in Polymer Science 170 (2025) 102029.doi:https://doi.org/10. 1016/j.progpolymsci.2025.102029

work page arXiv 2025
[6]

Y. Fang, H. Gao, Kinetic modeling for radical polymerization and depolymerization, Current Opin- ion in Chemical Engineering 49 (2025) 101152.doi:https://doi.org/10.1016/j.coche.2025. 101152

work page doi:10.1016/j.coche.2025 2025
[7]

N. E. Jackson, B. M. Savoie, Ten problems in polymer reactivity prediction, Macromolecules 58 (4) (2025) 1737–1754

2025
[8]

B. B. Noble, M. L. Coote, First principles modelling of free-radical polymerisation kinetics, Interna- tional Reviews in Physical Chemistry 32 (3) (2013) 467–513.doi:10.1080/0144235X.2013.797277

work page doi:10.1080/0144235x.2013.797277 2013
[9]

G. R. Jones, H. S. Wang, K. Parkatzidis, R. Whitfield, N. P. Truong, A. Anastasaki, Reversed controlled polymerization (rcp): depolymerization from well-defined polymers to monomers, Journal of the American Chemical Society 145 (18) (2023) 9898–9915.doi:10.1021/jacs.3c00589. 21

work page doi:10.1021/jacs.3c00589 2023
[10]

Wulkow, R

N. Wulkow, R. Telgmann, K.-D. Hungenberg, C. Schütte, M. Wulkow, Deterministic and stochastic parameter estimation for polymer reaction kinetics i: theory and simple examples, Macromolecular Theory and Simulations 30 (6) (2021) 2100017.doi:https://doi.org/10.1002/mats.202100017

work page doi:10.1002/mats.202100017 2021
[11]

A. M. Schweidtmann, A. D. Clayton, N. Holmes, E. Bradford, R. A. Bourne, A. A. Lapkin, Machine learning meets continuous flow chemistry: Automated optimization towards the pareto front of multiple objectives, Reaction Chemistry & Engineering 5 (5) (2020) 1075–1085.doi:https://doi. org/10.1016/j.cej.2018.07.031

work page doi:10.1016/j.cej.2018.07.031 2020
[12]

W. Ge, R. De Silva, Y. Fan, S. A. Sisson, M. H. Stenzel, Machine learning in polymer research, Advanced Materials 37 (11) (2025) 2413695.doi:https://doi.org/10.1002/adma.202413695

work page doi:10.1002/adma.202413695 2025
[13]

von Stosch, R

M. von Stosch, R. Oliveira, J. Peres, S. Feyo de Azevedo, Hybrid semi-parametric modeling in process systems engineering: Past, present and future, Computers & Chemical Engineering 60 (2014) 86–101.doi:https://doi.org/10.1016/j.compchemeng.2013.08.008

work page doi:10.1016/j.compchemeng.2013.08.008 2014
[14]

Y. Wang, Y. Fang, H. Zhou, H. Gao, A machine learning model for predicting the propaga- tion rate coefficient in free-radical polymerization, Molecules 29 (19) (2024) 4694.doi:10.3390/ molecules29194694

2024
[15]

Y. Shi, J. Wang, Q. Wang, Q. Jia, F. Yan, Z.-H. Luo, Y.-N. Zhou, Supervised machine learn- ing algorithms for predicting rate constants of ozone reaction with micropollutants, Industrial & Engineering Chemistry Research 61 (24) (2022) 8359–8367.doi:10.1021/acs.iecr.1c04697

work page doi:10.1021/acs.iecr.1c04697 2022
[16]

D. Li, Y. Ru, J. Liu, Gatboost: Mining graph attention networks-based important substructures of polymers for a better property prediction, Materials Today Communications 38 (2024) 107577. doi:https://doi.org/10.1016/j.mtcomm.2023.107577

work page doi:10.1016/j.mtcomm.2023.107577 2024
[17]

E. Inae, Y. Liu, Y. Zhu, J. Xu, G. Liu, R. Zhang, T. Luo, M. Jiang, Modeling Polymers with Neural Networks, American Chemical Society, 2025

2025
[18]

Mora-Mariano, A

D. Mora-Mariano, A. Flores-Tlacuahuac, I. Zapata-González, E. Saldívar-Guerra, Data-driven deep learning prediction of full molecular weight distribution in polymerization processes, The Cana- dian Journal of Chemical Engineering 103 (8) (2025) 3713–3725.arXiv:https://onlinelibrary. wiley.com/doi/pdf/10.1002/cjce.25635,doi:https://doi.org/10.1002/cjce.25635

work page doi:10.1002/cjce.25635 2025
[19]

Bardooli, Y

A. Bardooli, Y. Dong, C. Georgakis, Data-driven process modeling and optimization aided by ma- terial and energy balances: The case of a batch polymerization process, IFAC-PapersOnLine 54 (3) (2021) 1–6, 16th IFAC Symposium on Advanced Control of Chemical Processes ADCHEM 2021. doi:https://doi.org/10.1016/j.ifacol.2021.08.209

work page doi:10.1016/j.ifacol.2021.08.209 2021
[20]

Z. J. Liew, Z. Elkhaiary, A. A. Lapkin, Parameter efficient multi-model vision assistant for polymer solvation behaviour inference, npj Computational Materials 11 (1) (2025) 161

2025
[21]

G. Mogk, T. Mrziglod, A. Schuppert, Application of hybrid models in chemical industry, in: J. Grievink, J. van Schijndel (Eds.), European Symposium on Computer Aided Process Engineering- 12, Vol. 10 of Computer Aided Chemical Engineering, Elsevier, 2002, pp. 931–936.doi:https: //doi.org/10.1016/S1570-7946(02)80183-3

work page doi:10.1016/s1570-7946(02)80183-3 2002
[22]

Kahrs, W

O. Kahrs, W. Marquardt, Incremental identification of hybrid process models, Computers & Chem- ical Engineering 32 (4) (2008) 694–705, festschrift devoted to Rex Reklaitis on his 65th Birthday. doi:https://doi.org/10.1016/j.compchemeng.2007.02.014. 22

work page doi:10.1016/j.compchemeng.2007.02.014 2008
[23]

D. C. Psichogios, L. H. Ungar, A hybrid neural network-first principles approach to process mod- eling, AIChE Journal 38 (10) (1992) 1499–1511.arXiv:https://aiche.onlinelibrary.wiley. com/doi/pdf/10.1002/aic.690381003,doi:https://doi.org/10.1002/aic.690381003

work page doi:10.1002/aic.690381003 1992
[24]

Merkelbach, A

K. Merkelbach, A. M. Schweidtmann, Y. Müller, P. Schwoebel, A. Mhamdi, A. Mitsos, A. Schuppert, T. Mrziglod, S. Schneckener, Hybridml: Open source platform for hybrid modeling, Computers & Chemical Engineering 160 (2022) 107736.doi:https://doi.org/10.1016/j.compchemeng.2022. 107736

work page doi:10.1016/j.compchemeng.2022 2022
[25]

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440.doi:https://doi.org/10.1038/ s42254-021-00314-5

2021
[26]

D. J. Audus, J. J. de Pablo, Polymer informatics: opportunities and challenges, ACS macro letters 6 (10) (2017) 1078–1082.doi:10.1021/acsmacrolett.7b00228

work page doi:10.1021/acsmacrolett.7b00228 2017
[27]

Lin, H.-Y

D. Lin, H.-Y. Yu, Deep learning and inverse discovery of polymer self-consistent field theory inspired by physics-informed neural networks, Physical Review E 106 (1) (2022) 014503.doi:https://doi. org/10.1103/PhysRevE.106.014503

work page doi:10.1103/physreve.106.014503 2022
[28]

J. G. Ethier, D. J. Audus, D. C. Ryan, R. A. Vaia, Integrating theory with machine learning for predicting polymer solution phase behavior, Giant 15 (2023) 100171.doi:https://doi.org/10. 1016/j.giant.2023.100171

work page arXiv 2023
[29]

Y. Li, Y. Lin, Z. Mai, Q. Deng, Z. Ju, B. Cao, W. Li, Theory-guided machine learning for strength prediction of polymers at different temperatures and strain rates with physical information, Mate- rials Today Communications 48 (2025) 113519.doi:https://doi.org/10.1016/j.mtcomm.2025. 113519

work page doi:10.1016/j.mtcomm.2025 2025
[30]

Velioglu, S

M. Velioglu, S. Zhai, S. Rupprecht, A. Mitsos, A. Jupke, M. Dahmen, Physics-informed neural networks for dynamic process operations with limited physical knowledge and data, Computers & Chemical Engineering 192 (2025) 108899.doi:https://doi.org/10.1016/j.compchemeng.2024. 108899

work page doi:10.1016/j.compchemeng.2024 2025
[31]

D. J. Audus, A. McDannald, B. DeCost, Leveraging theory for enhanced machine learning, ACS Macro Letters 11 (9) (2022) 1117–1122, pMID: 36018715.doi:10.1021/acsmacrolett.2c00369

work page doi:10.1021/acsmacrolett.2c00369 2022
[32]

Zobeiry, A

N. Zobeiry, A. Poursartip, Theory-guided machine learning for process simulation of advanced com- posites, arXiv preprint arXiv:2103.16010 (2021).doi:https://doi.org/10.48550/arXiv.2103. 16010

work page doi:10.48550/arxiv.2103 2021
[33]

R. T. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equations, Advances in neural information processing systems 31 (2018)

2018
[34]

Massaroli, M

S. Massaroli, M. Poli, J. Park, A. Yamashita, H. Asama, Dissecting neural odes, in: H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, H. Lin (Eds.), Advances in Neural Information Processing Systems, Vol. 33, Curran Associates, Inc., 2020, pp. 3952–3963

2020
[35]

Sorourifar, Y

F. Sorourifar, Y. Peng, I. Castillo, L. Bui, J. Venegas, J. A. Paulson, Physics-enhanced neural ordinary differential equations: Application to industrial chemical reaction systems, Industrial & Engineering Chemistry Research 62 (38) (2023) 15563–15577. 23

2023
[36]

H. Yan, J. Du, V. Y. F. Tan, J. Feng, On robustness of neural ordinary differential equations (2022). arXiv:1910.05513. URLhttps://arxiv.org/abs/1910.05513

work page arXiv 2022
[37]

G. T. Russell, The kinetics of free-radical polymerization: Fundamental aspects, Australian Journal of Chemistry 55 (7) (2002) 399–414.arXiv:https://connectsci.au/ch/article-pdf/55/7/399/ 104682/ch02114.pdf,doi:10.1071/CH02114

work page doi:10.1071/ch02114 2002
[38]

O.Walz, C.Marks, J.Viell, A.Mitsos, Systematicapproachformodelingreactionnetworksinvolving equilibrium and kinetically-limited reaction steps, Computers & Chemical Engineering 98 (2017) 143–153.doi:https://doi.org/10.1016/j.compchemeng.2016.12.014

work page doi:10.1016/j.compchemeng.2016.12.014 2017
[39]

Matyjaszewski, T

K. Matyjaszewski, T. P. Davis, et al., Handbook of radical polymerization, Vol. 922, Wiley Online Library, 2002

2002
[40]

Guerrero-Santos, E

R. Guerrero-Santos, E. Saldívar-Guerra, I. Zapata-González, J. Bonilla-Cruz, E. Vivaldo-Lima, Free-Radical Polymerization, John Wiley & Sons, Ltd, 2025, Ch. 3, pp. 65–95.arXiv: https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781119820123.ch3,doi:https://doi. org/10.1002/9781119820123.ch3

work page doi:10.1002/9781119820123.ch3 2025
[41]

Ehlers, J

F. Ehlers, J. Barth, P. Vana, Kinetics and thermodynamics of radical polymerization, in: Funda- mentals of Controlled/Living Radical Polymerization, The Royal Society of Chemistry, 2013, pp. 1–59.doi:10.1039/9781849737425-00001

work page doi:10.1039/9781849737425-00001 2013
[42]

G. Moad, D. H. Solomon, The chemistry of radical polymerization, Elsevier, 2005

2005
[43]

Butté, G

A. Butté, G. Storti, M. Morbidelli, Evaluation of the chain length distribution in free-radical poly- merization, 1. bulk polymerization, Macromolecular theory and simulations 11 (1) (2002) 22–36

2002
[44]

Butté, G

A. Butté, G. Storti, M. Morbidelli, Evaluation of the chain length distribution in free-radical poly- merization, 2. emulsion polymerization, Macromolecular theory and simulations 11 (1) (2002) 37–52

2002
[45]

Sanderse, P

B. Sanderse, P. Stinis, R. Maulik, S. E. Ahmed, Scientific machine learning for closure models in multiscale problems: a review (2024).arXiv:2403.02913. URLhttps://arxiv.org/abs/2403.02913

work page arXiv 2024
[46]

P. E. Kloeden, M. Rasmussen, Nonautonomous dynamical systems, no. 176, American Mathematical Soc., 2011

2011
[47]

X. Wang, E. K. Blum, Discrete-time versus continuous-time models of neural networks, Jour- nal of Computer and System Sciences 45 (1) (1992) 1–19.doi:https://doi.org/10.1016/ 0022-0000(92)90038-K

1992
[48]

Massaroli, M

S. Massaroli, M. Poli, J. Park, A. Yamashita, H. Asama, Dissecting neural odes, Advances in neural information processing systems 33 (2020) 3952–3963

2020
[49]

J. D. Tan, B. Ramalingam, S. L. Wong, J. Cheng, Y.-F. Lim, V. Chellappan, S. A. Khan, J. Ku- mar, K. Hippalgaonkar, Machine learning predicts conversion and molecular weight distributions in computer controlled polymerization, Journal of Chemical Information and Modeling (2022)

2022
[50]

J. R. Richards, J. P. Congalidis, Measurement and control of polymerization reactors, Computers & Chemical Engineering 30 (10) (2006) 1447–1463, papers form Chemical Process Control VII. doi:https://doi.org/10.1016/j.compchemeng.2006.05.021. 24

work page doi:10.1016/j.compchemeng.2006.05.021 2006
[51]

Paszke, S

A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high-performance deep learning library, Advances in neural information processing systems 32 (2019)

2019
[52]

R. T. Q. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equa- tions, Advances in neural information processing systems 31 (2018)

2018
[53]

Virtanen, R

P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, et al., Scipy 1.0: fundamental algorithms for scientific computing in python, Nature methods 17 (3) (2020) 261–272.doi:https://doi.org/10.1038/ s41592-019-0686-2

2020
[54]

Decoupled Weight Decay Regularization

I. Loshchilov, F. Hutter, Decoupled weight decay regularization (2019).arXiv:1711.05101. URLhttps://arxiv.org/abs/1711.05101 25

work page internal anchor Pith review Pith/arXiv arXiv 2019

[1] [1]

Ohshima, M

M. Ohshima, M. Tanigaki, Quality control of polymer production processes, Journal of Process Control 10 (2) (2000) 135–148.doi:https://doi.org/10.1016/S0959-1524(99)00042-6

work page doi:10.1016/s0959-1524(99)00042-6 2000

[2] [2]

Karuppusamy, R

M. Karuppusamy, R. Thirumalaisamy, S. Palanisamy, S. Nagamalai, E. E. S. Massoud, N. Ayrilmis, A review of machine learning applications in polymer composites: advancements, challenges, and future prospects, Journal of Materials Chemistry A 13 (22) (2025) 16290–16308.doi:10.1039/ D5TA00982K

2025

[3] [3]

J. M. Asua, Emulsion polymerization: From fundamental mechanisms to process developments, Journal of Polymer Science Part A: Polymer Chemistry 42 (5) (2004) 1025–1041.doi:https: //doi.org/10.1002/pola.11096

work page doi:10.1002/pola.11096 2004

[4] [5]

Farajzadehahary, S

K. Farajzadehahary, S. Hamzehlou, N. Ballard, Adding machine learning to the polymer reaction engineering toolbox, Progress in Polymer Science 170 (2025) 102029.doi:https://doi.org/10. 1016/j.progpolymsci.2025.102029

work page arXiv 2025

[5] [6]

Y. Fang, H. Gao, Kinetic modeling for radical polymerization and depolymerization, Current Opin- ion in Chemical Engineering 49 (2025) 101152.doi:https://doi.org/10.1016/j.coche.2025. 101152

work page doi:10.1016/j.coche.2025 2025

[6] [7]

N. E. Jackson, B. M. Savoie, Ten problems in polymer reactivity prediction, Macromolecules 58 (4) (2025) 1737–1754

2025

[7] [8]

B. B. Noble, M. L. Coote, First principles modelling of free-radical polymerisation kinetics, Interna- tional Reviews in Physical Chemistry 32 (3) (2013) 467–513.doi:10.1080/0144235X.2013.797277

work page doi:10.1080/0144235x.2013.797277 2013

[8] [9]

G. R. Jones, H. S. Wang, K. Parkatzidis, R. Whitfield, N. P. Truong, A. Anastasaki, Reversed controlled polymerization (rcp): depolymerization from well-defined polymers to monomers, Journal of the American Chemical Society 145 (18) (2023) 9898–9915.doi:10.1021/jacs.3c00589. 21

work page doi:10.1021/jacs.3c00589 2023

[9] [10]

Wulkow, R

N. Wulkow, R. Telgmann, K.-D. Hungenberg, C. Schütte, M. Wulkow, Deterministic and stochastic parameter estimation for polymer reaction kinetics i: theory and simple examples, Macromolecular Theory and Simulations 30 (6) (2021) 2100017.doi:https://doi.org/10.1002/mats.202100017

work page doi:10.1002/mats.202100017 2021

[10] [11]

A. M. Schweidtmann, A. D. Clayton, N. Holmes, E. Bradford, R. A. Bourne, A. A. Lapkin, Machine learning meets continuous flow chemistry: Automated optimization towards the pareto front of multiple objectives, Reaction Chemistry & Engineering 5 (5) (2020) 1075–1085.doi:https://doi. org/10.1016/j.cej.2018.07.031

work page doi:10.1016/j.cej.2018.07.031 2020

[11] [12]

W. Ge, R. De Silva, Y. Fan, S. A. Sisson, M. H. Stenzel, Machine learning in polymer research, Advanced Materials 37 (11) (2025) 2413695.doi:https://doi.org/10.1002/adma.202413695

work page doi:10.1002/adma.202413695 2025

[12] [13]

von Stosch, R

M. von Stosch, R. Oliveira, J. Peres, S. Feyo de Azevedo, Hybrid semi-parametric modeling in process systems engineering: Past, present and future, Computers & Chemical Engineering 60 (2014) 86–101.doi:https://doi.org/10.1016/j.compchemeng.2013.08.008

work page doi:10.1016/j.compchemeng.2013.08.008 2014

[13] [14]

Y. Wang, Y. Fang, H. Zhou, H. Gao, A machine learning model for predicting the propaga- tion rate coefficient in free-radical polymerization, Molecules 29 (19) (2024) 4694.doi:10.3390/ molecules29194694

2024

[14] [15]

Y. Shi, J. Wang, Q. Wang, Q. Jia, F. Yan, Z.-H. Luo, Y.-N. Zhou, Supervised machine learn- ing algorithms for predicting rate constants of ozone reaction with micropollutants, Industrial & Engineering Chemistry Research 61 (24) (2022) 8359–8367.doi:10.1021/acs.iecr.1c04697

work page doi:10.1021/acs.iecr.1c04697 2022

[15] [16]

D. Li, Y. Ru, J. Liu, Gatboost: Mining graph attention networks-based important substructures of polymers for a better property prediction, Materials Today Communications 38 (2024) 107577. doi:https://doi.org/10.1016/j.mtcomm.2023.107577

work page doi:10.1016/j.mtcomm.2023.107577 2024

[16] [17]

E. Inae, Y. Liu, Y. Zhu, J. Xu, G. Liu, R. Zhang, T. Luo, M. Jiang, Modeling Polymers with Neural Networks, American Chemical Society, 2025

2025

[17] [18]

Mora-Mariano, A

D. Mora-Mariano, A. Flores-Tlacuahuac, I. Zapata-González, E. Saldívar-Guerra, Data-driven deep learning prediction of full molecular weight distribution in polymerization processes, The Cana- dian Journal of Chemical Engineering 103 (8) (2025) 3713–3725.arXiv:https://onlinelibrary. wiley.com/doi/pdf/10.1002/cjce.25635,doi:https://doi.org/10.1002/cjce.25635

work page doi:10.1002/cjce.25635 2025

[18] [19]

Bardooli, Y

A. Bardooli, Y. Dong, C. Georgakis, Data-driven process modeling and optimization aided by ma- terial and energy balances: The case of a batch polymerization process, IFAC-PapersOnLine 54 (3) (2021) 1–6, 16th IFAC Symposium on Advanced Control of Chemical Processes ADCHEM 2021. doi:https://doi.org/10.1016/j.ifacol.2021.08.209

work page doi:10.1016/j.ifacol.2021.08.209 2021

[19] [20]

Z. J. Liew, Z. Elkhaiary, A. A. Lapkin, Parameter efficient multi-model vision assistant for polymer solvation behaviour inference, npj Computational Materials 11 (1) (2025) 161

2025

[20] [21]

G. Mogk, T. Mrziglod, A. Schuppert, Application of hybrid models in chemical industry, in: J. Grievink, J. van Schijndel (Eds.), European Symposium on Computer Aided Process Engineering- 12, Vol. 10 of Computer Aided Chemical Engineering, Elsevier, 2002, pp. 931–936.doi:https: //doi.org/10.1016/S1570-7946(02)80183-3

work page doi:10.1016/s1570-7946(02)80183-3 2002

[21] [22]

Kahrs, W

O. Kahrs, W. Marquardt, Incremental identification of hybrid process models, Computers & Chem- ical Engineering 32 (4) (2008) 694–705, festschrift devoted to Rex Reklaitis on his 65th Birthday. doi:https://doi.org/10.1016/j.compchemeng.2007.02.014. 22

work page doi:10.1016/j.compchemeng.2007.02.014 2008

[22] [23]

D. C. Psichogios, L. H. Ungar, A hybrid neural network-first principles approach to process mod- eling, AIChE Journal 38 (10) (1992) 1499–1511.arXiv:https://aiche.onlinelibrary.wiley. com/doi/pdf/10.1002/aic.690381003,doi:https://doi.org/10.1002/aic.690381003

work page doi:10.1002/aic.690381003 1992

[23] [24]

Merkelbach, A

K. Merkelbach, A. M. Schweidtmann, Y. Müller, P. Schwoebel, A. Mhamdi, A. Mitsos, A. Schuppert, T. Mrziglod, S. Schneckener, Hybridml: Open source platform for hybrid modeling, Computers & Chemical Engineering 160 (2022) 107736.doi:https://doi.org/10.1016/j.compchemeng.2022. 107736

work page doi:10.1016/j.compchemeng.2022 2022

[24] [25]

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440.doi:https://doi.org/10.1038/ s42254-021-00314-5

2021

[25] [26]

D. J. Audus, J. J. de Pablo, Polymer informatics: opportunities and challenges, ACS macro letters 6 (10) (2017) 1078–1082.doi:10.1021/acsmacrolett.7b00228

work page doi:10.1021/acsmacrolett.7b00228 2017

[26] [27]

Lin, H.-Y

D. Lin, H.-Y. Yu, Deep learning and inverse discovery of polymer self-consistent field theory inspired by physics-informed neural networks, Physical Review E 106 (1) (2022) 014503.doi:https://doi. org/10.1103/PhysRevE.106.014503

work page doi:10.1103/physreve.106.014503 2022

[27] [28]

J. G. Ethier, D. J. Audus, D. C. Ryan, R. A. Vaia, Integrating theory with machine learning for predicting polymer solution phase behavior, Giant 15 (2023) 100171.doi:https://doi.org/10. 1016/j.giant.2023.100171

work page arXiv 2023

[28] [29]

Y. Li, Y. Lin, Z. Mai, Q. Deng, Z. Ju, B. Cao, W. Li, Theory-guided machine learning for strength prediction of polymers at different temperatures and strain rates with physical information, Mate- rials Today Communications 48 (2025) 113519.doi:https://doi.org/10.1016/j.mtcomm.2025. 113519

work page doi:10.1016/j.mtcomm.2025 2025

[29] [30]

Velioglu, S

M. Velioglu, S. Zhai, S. Rupprecht, A. Mitsos, A. Jupke, M. Dahmen, Physics-informed neural networks for dynamic process operations with limited physical knowledge and data, Computers & Chemical Engineering 192 (2025) 108899.doi:https://doi.org/10.1016/j.compchemeng.2024. 108899

work page doi:10.1016/j.compchemeng.2024 2025

[30] [31]

D. J. Audus, A. McDannald, B. DeCost, Leveraging theory for enhanced machine learning, ACS Macro Letters 11 (9) (2022) 1117–1122, pMID: 36018715.doi:10.1021/acsmacrolett.2c00369

work page doi:10.1021/acsmacrolett.2c00369 2022

[31] [32]

Zobeiry, A

N. Zobeiry, A. Poursartip, Theory-guided machine learning for process simulation of advanced com- posites, arXiv preprint arXiv:2103.16010 (2021).doi:https://doi.org/10.48550/arXiv.2103. 16010

work page doi:10.48550/arxiv.2103 2021

[32] [33]

R. T. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equations, Advances in neural information processing systems 31 (2018)

2018

[33] [34]

Massaroli, M

S. Massaroli, M. Poli, J. Park, A. Yamashita, H. Asama, Dissecting neural odes, in: H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, H. Lin (Eds.), Advances in Neural Information Processing Systems, Vol. 33, Curran Associates, Inc., 2020, pp. 3952–3963

2020

[34] [35]

Sorourifar, Y

F. Sorourifar, Y. Peng, I. Castillo, L. Bui, J. Venegas, J. A. Paulson, Physics-enhanced neural ordinary differential equations: Application to industrial chemical reaction systems, Industrial & Engineering Chemistry Research 62 (38) (2023) 15563–15577. 23

2023

[35] [36]

H. Yan, J. Du, V. Y. F. Tan, J. Feng, On robustness of neural ordinary differential equations (2022). arXiv:1910.05513. URLhttps://arxiv.org/abs/1910.05513

work page arXiv 2022

[36] [37]

G. T. Russell, The kinetics of free-radical polymerization: Fundamental aspects, Australian Journal of Chemistry 55 (7) (2002) 399–414.arXiv:https://connectsci.au/ch/article-pdf/55/7/399/ 104682/ch02114.pdf,doi:10.1071/CH02114

work page doi:10.1071/ch02114 2002

[37] [38]

O.Walz, C.Marks, J.Viell, A.Mitsos, Systematicapproachformodelingreactionnetworksinvolving equilibrium and kinetically-limited reaction steps, Computers & Chemical Engineering 98 (2017) 143–153.doi:https://doi.org/10.1016/j.compchemeng.2016.12.014

work page doi:10.1016/j.compchemeng.2016.12.014 2017

[38] [39]

Matyjaszewski, T

K. Matyjaszewski, T. P. Davis, et al., Handbook of radical polymerization, Vol. 922, Wiley Online Library, 2002

2002

[39] [40]

Guerrero-Santos, E

R. Guerrero-Santos, E. Saldívar-Guerra, I. Zapata-González, J. Bonilla-Cruz, E. Vivaldo-Lima, Free-Radical Polymerization, John Wiley & Sons, Ltd, 2025, Ch. 3, pp. 65–95.arXiv: https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781119820123.ch3,doi:https://doi. org/10.1002/9781119820123.ch3

work page doi:10.1002/9781119820123.ch3 2025

[40] [41]

Ehlers, J

F. Ehlers, J. Barth, P. Vana, Kinetics and thermodynamics of radical polymerization, in: Funda- mentals of Controlled/Living Radical Polymerization, The Royal Society of Chemistry, 2013, pp. 1–59.doi:10.1039/9781849737425-00001

work page doi:10.1039/9781849737425-00001 2013

[41] [42]

G. Moad, D. H. Solomon, The chemistry of radical polymerization, Elsevier, 2005

2005

[42] [43]

Butté, G

A. Butté, G. Storti, M. Morbidelli, Evaluation of the chain length distribution in free-radical poly- merization, 1. bulk polymerization, Macromolecular theory and simulations 11 (1) (2002) 22–36

2002

[43] [44]

Butté, G

A. Butté, G. Storti, M. Morbidelli, Evaluation of the chain length distribution in free-radical poly- merization, 2. emulsion polymerization, Macromolecular theory and simulations 11 (1) (2002) 37–52

2002

[44] [45]

Sanderse, P

B. Sanderse, P. Stinis, R. Maulik, S. E. Ahmed, Scientific machine learning for closure models in multiscale problems: a review (2024).arXiv:2403.02913. URLhttps://arxiv.org/abs/2403.02913

work page arXiv 2024

[45] [46]

P. E. Kloeden, M. Rasmussen, Nonautonomous dynamical systems, no. 176, American Mathematical Soc., 2011

2011

[46] [47]

X. Wang, E. K. Blum, Discrete-time versus continuous-time models of neural networks, Jour- nal of Computer and System Sciences 45 (1) (1992) 1–19.doi:https://doi.org/10.1016/ 0022-0000(92)90038-K

1992

[47] [48]

Massaroli, M

S. Massaroli, M. Poli, J. Park, A. Yamashita, H. Asama, Dissecting neural odes, Advances in neural information processing systems 33 (2020) 3952–3963

2020

[48] [49]

J. D. Tan, B. Ramalingam, S. L. Wong, J. Cheng, Y.-F. Lim, V. Chellappan, S. A. Khan, J. Ku- mar, K. Hippalgaonkar, Machine learning predicts conversion and molecular weight distributions in computer controlled polymerization, Journal of Chemical Information and Modeling (2022)

2022

[49] [50]

J. R. Richards, J. P. Congalidis, Measurement and control of polymerization reactors, Computers & Chemical Engineering 30 (10) (2006) 1447–1463, papers form Chemical Process Control VII. doi:https://doi.org/10.1016/j.compchemeng.2006.05.021. 24

work page doi:10.1016/j.compchemeng.2006.05.021 2006

[50] [51]

Paszke, S

A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high-performance deep learning library, Advances in neural information processing systems 32 (2019)

2019

[51] [52]

R. T. Q. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equa- tions, Advances in neural information processing systems 31 (2018)

2018

[52] [53]

Virtanen, R

P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, et al., Scipy 1.0: fundamental algorithms for scientific computing in python, Nature methods 17 (3) (2020) 261–272.doi:https://doi.org/10.1038/ s41592-019-0686-2

2020

[53] [54]

Decoupled Weight Decay Regularization

I. Loshchilov, F. Hutter, Decoupled weight decay regularization (2019).arXiv:1711.05101. URLhttps://arxiv.org/abs/1711.05101 25

work page internal anchor Pith review Pith/arXiv arXiv 2019