pith. sign in

arxiv: 2605.20358 · v1 · pith:KIC6MB5Nnew · submitted 2026-05-19 · ❄️ cond-mat.mtrl-sci · cond-mat.dis-nn· cond-mat.mes-hall

Modeling phase separation in polymer-derived carbonitride ceramics through extended machine learning molecular dynamics

Fabien Mortier , Sylvian Cadars , Olivier Masson , Mauro Boero , Guido Ori , Yun Wang , Samuel Bernard , Assil Bouzid This is my paper

Pith reviewed 2026-05-21 07:24 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cond-mat.dis-nncond-mat.mes-hall
keywords polymer-derived ceramicssilicon carbonitridephase separationmachine learning interatomic potentialmolecular dynamicsgraphene-like domainsamorphous materialsthermal processing
0
0 comments X

The pith

Machine learning molecular dynamics reveals carbon nucleating into graphene-like sheets within the SiCN ceramic matrix during heating.

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

This paper develops a machine learning interatomic potential for Si-C-N-H systems trained on more than 9000 configurations that include amorphous structures, high-temperature states, surfaces, and crystals. The potential supports molecular dynamics simulations of 8000-atom models that track the evolution of polymer-derived silicon carbonitride during thermal processing. These simulations show carbon domains progressively separating from the amorphous matrix to form distinct graphene-like sheets while the surrounding ceramic network remains connected. The resulting structures match experimental atomic pair distribution functions and trace the transformation through rearrangements of defective five- and seven-membered carbon rings into stable six-membered rings.

Core claim

The central discovery is a phase separation process in which carbon domains nucleate from the amorphous SiCN matrix during thermal treatment, forming graphene-like sheets while the integrity of the ceramic network is preserved. Defective five- and seven-membered carbon rings mediate the conversion to stable six-membered aromatic structures. The simulated atomic configurations reproduce experimental pair distribution functions with high fidelity, supplying atomic-scale explanations for the combination of ceramic thermal stability and graphitic features in these materials.

What carries the argument

A machine learning interatomic potential trained on a diversified database of over 9000 configurations that enables large-scale molecular dynamics simulations of 8000-atom SiCNH systems at high temperature.

If this is right

  • The method supplies atomic-level mechanisms for the thermal stability and mixed properties of polymer-derived ceramics.
  • Defective carbon rings are identified as the structural intermediates that enable the transition to ordered graphitic domains.
  • The same potential framework can be applied to other complex amorphous multicomponent systems at experimentally relevant sizes.
  • Microscopic pathways for structural transformation during processing are now accessible for direct comparison with experiment.

Where Pith is reading between the lines

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

  • Processing temperatures or times could be adjusted to control the size and connectivity of the emerging carbon sheets for targeted electrical or mechanical performance.
  • The same simulation approach might be extended to predict how dopants or different polymer precursors alter the phase-separation route.
  • Longer simulation timescales could reveal whether the graphene-like domains continue to grow or stabilize once formed.

Load-bearing premise

The trained machine learning potential reproduces the energies and dynamics of carbon nucleation and ring changes in large systems without large extrapolation errors or missing interactions.

What would settle it

A simulation run at the same temperature and scale that produces no carbon domain nucleation or that yields atomic pair distribution functions clearly different from experimental measurements would falsify the reported phase-separation mechanism.

Figures

Figures reproduced from arXiv: 2605.20358 by Assil Bouzid, Fabien Mortier, Guido Ori, Mauro Boero, Olivier Masson, Samuel Bernard, Sylvian Cadars, Yun Wang.

Figure 1
Figure 1. Figure 1: FIG. 1: Multidimensional scaling 2D projection of the training database using Valle-Oganov fingerprints [ [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Evolution of the loss on the validation set during the training process. The figure shows the two stages of [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Comparison of MLIP-predicted energies and forces with DFT reference values for the test set. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: a) Thermal cycles applied to the 8000-atom models using the MLIP. The extended high-temperature [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Partial pair distribution functions for the 8000-atom models. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: (Top) Configurations of C-C bonds in the final snapshot of models SiCNH8000-Random, SiCNH8000-CNB, [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Evolution of the proportion of carbon atoms in different environments (single, dimers, linear chains, [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: (a-f) Atomic configurations illustrating the mechanism of aromatic ring formation from isolated carbon [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
read the original abstract

Polymer-derived ceramics combine the thermal stability of ceramics with the versatile properties of carbon domains, but modeling their atomic-scale evolution during processing remains elusive due to the limitations of traditional computational methods. To address this issue, here we develop and apply a machine learning interatomic potential for silicon carbonitride-based (Si-C-N-H) systems, trained on a diversified database of over 9000 configurations -including amorphous models, high-temperature states, surfaces, and crystal structure predictions - to capture the full complexity of these materials. This potential enables large-scale molecular dynamics simulations of 8000-atom systems revealing the atomic-scale evolution of the polymer-derived ceramic during thermal treatment. A key result of this work is the occurrence of a phase separation where carbon domains progressively nucleate from the amorphous SiCN matrix during thermal processing, forming distinct graphene-like sheets while preserving the integrity of the ceramic network. The resulting models reproduce the experimental atomic pair distribution functions with exceptional fidelity, validating our approach and providing microscopic explanations for the material unique combination of ceramic and graphitic properties. In this process, defective 5- and/or 7-member carbon rings, mediate the transformation to stable 6-member aromatic structures. These findings offer new atomic-scale insights into the thermal stability and structural transformation pathways of polymer-derived ceramics, while our methodology opens avenues for studying complex amorphous systems with first-principles accuracy at experimentally relevant scales.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 2 minor

Summary. The paper develops a machine learning interatomic potential for Si-C-N-H systems trained on a diversified database of over 9000 configurations (including amorphous, high-temperature, surface, and crystal structures). It applies this potential to large-scale MD simulations of 8000-atom polymer-derived SiCN models during thermal processing. The central claim is that carbon domains nucleate from the amorphous matrix to form graphene-like sheets via mediation by defective 5- and 7-membered rings that rearrange into stable 6-membered aromatic structures, while the final atomic configurations reproduce experimental pair distribution functions with high fidelity.

Significance. If the dynamical pathway is faithfully reproduced, the work provides valuable atomic-scale mechanistic insight into phase separation and carbon domain formation in polymer-derived ceramics, explaining their hybrid ceramic-graphitic properties at experimentally relevant scales. The scale-up to 8000-atom systems and reproduction of experimental PDFs are notable strengths that could inform processing strategies.

major comments (1)
  1. [MD simulation results and validation section] The load-bearing claim is that the ML potential correctly drives carbon nucleation and the 5-/7- to 6-membered ring conversion during high-T annealing. However, validation is reported only via reproduction of experimental PDFs for the final structures (Abstract and results on MD trajectories). PDF agreement constrains equilibrium pair correlations but does not test relative energies or barriers of the defective-ring intermediates identified as mediating the transformation, leaving open the possibility that the observed pathway is an extrapolation artifact rather than a physically faithful mechanism.
minor comments (2)
  1. [Methods] The training database is described at a high level ('diversified database of over 9000 configurations'); more detail on the distribution of high-temperature states and how extrapolation risk was quantified for the 8000-atom runs would strengthen the methods.
  2. [Results] Error bars or uncertainty estimates on the PDF comparisons and on the reported ring statistics during annealing are not mentioned; adding these would improve clarity of the validation.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive review of our manuscript. We address the major comment regarding validation of the dynamical pathway below, and we outline revisions that will strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [MD simulation results and validation section] The load-bearing claim is that the ML potential correctly drives carbon nucleation and the 5-/7- to 6-membered ring conversion during high-T annealing. However, validation is reported only via reproduction of experimental PDFs for the final structures (Abstract and results on MD trajectories). PDF agreement constrains equilibrium pair correlations but does not test relative energies or barriers of the defective-ring intermediates identified as mediating the transformation, leaving open the possibility that the observed pathway is an extrapolation artifact rather than a physically faithful mechanism.

    Authors: We agree that reproduction of experimental PDFs validates the final atomic configurations but does not directly probe the relative energies or barriers associated with the 5-/7- to 6-membered ring rearrangements. Our ML potential was trained on a diversified database of over 9000 configurations that explicitly includes high-temperature amorphous states, surfaces, and crystal structures, which encompass a broad range of carbon ring environments and defective motifs. This training strategy is intended to ensure faithful description of the relevant energetics. Nevertheless, to directly address the concern about possible extrapolation artifacts, the revised manuscript will include new analysis: we will report relative energies (computed with both the ML potential and reference DFT) for representative small-model systems containing 5-, 6-, and 7-membered carbon rings and their interconversions. These additional checks will confirm that the observed mediation pathway is consistent with the underlying potential energy surface learned from the training data. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external training data and experimental validation

full rationale

The paper trains a machine learning interatomic potential on an external diversified database of over 9000 configurations (including amorphous models and high-temperature states) and then performs large-scale MD simulations on 8000-atom systems. The reported phase separation and ring rearrangements emerge as simulation outcomes, with final structures validated against independent experimental PDFs. No equations, fitted parameters, or self-citations reduce the central claim to the inputs by construction; the workflow remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the ML potential trained on the stated database accurately captures high-temperature dynamics and that the observed phase separation is not an artifact of training-data selection or model limitations.

free parameters (1)
  • ML interatomic potential hyperparameters
    Neural-network weights and architecture choices fitted during training on the 9000-configuration database.
axioms (1)
  • domain assumption The diversified training database of over 9000 configurations sufficiently samples the relevant configuration space for amorphous, high-temperature, and surface states in Si-C-N-H systems.
    Invoked to justify that the potential can be applied to large-scale thermal-processing simulations without major extrapolation errors.

pith-pipeline@v0.9.0 · 5811 in / 1453 out tokens · 43339 ms · 2026-05-21T07:24:40.496184+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

78 extracted references · 78 canonical work pages

  1. [1]

    Configurations derived from first-principles molecular dynamics The SiCNH-periodic class contains configurations extracted from FPMD trajectories of periodic models with com- position Si32C25N24H19. This class includes configurations from seven models presented in our previous work [30] in addition to three new models produced with the purpose of increasi...

    work page

  2. [2]

    Initial configurations were obtained from the work of Deringer and Csányi

    Amorphous Carbon and Silicon configurations The Amorphous-Carbon-200 and Amorphous-Carbon-250 classes incorporate amorphous carbon structures at den- sities of 2.00 and 2.50 g/cm3, respectively. Initial configurations were obtained from the work of Deringer and Csányi

    work page

  3. [3]

    Each initial amorphous carbon configuration was subjected to a short thermal cycle of about 20 ps at T = 300 K, from which 249 configurations were extracted from each model

    where melt-quenched models were produced at different densities, resulting in models with carbon sp, sp2, and sp3 hybridizations. Each initial amorphous carbon configuration was subjected to a short thermal cycle of about 20 ps at T = 300 K, from which 249 configurations were extracted from each model. Similarly, for the Amorphous-Silicon class, we collec...

    work page

  4. [4]

    Crystal Structure Predictions The Crystal-Structure-Prediction class contains 1165 configurations generated using the USPEX program [43–45] for various Si-C-N-H compositions and densities. USPEX is a crystal structure prediction (CSP) program employing evolutionary algorithm to generate, generation after generation, a diverse set of candidate structures f...

    work page 2000

  5. [5]

    Figure 5 presents the partial PDFs for the four 8000-atom models excluding hydrogen-containing pairs, which provide only very small contributions to the total PDF

    Partial pair distribution functions The models being validated, we now focus on their structural characterization. Figure 5 presents the partial PDFs for the four 8000-atom models excluding hydrogen-containing pairs, which provide only very small contributions to the total PDF. Before examining the individual partial PDFs, we first point that most major p...

    work page

  6. [6]

    We find that, in the four models, between 87% and 90% of silicon atoms are 4-fold coordinated, as expected from a Si-based tetrahedral network

    Local atomic environments Table VI shows the dominant Si local environments in the four 8000-atom models and Tables S4, S5, S6 and S7 in the supplementary materials, provide the complete list of local environments for all the models. We find that, in the four models, between 87% and 90% of silicon atoms are 4-fold coordinated, as expected from a Si-based ...

    work page

  7. [7]

    single" carbon atoms only bonded to silicon, nitrogen, and/or hydrogen, i.e. completely embedded in the SiCN network. The second category, labelled

    Evolution of C motifs Figure S7 in the supplementary materials shows snapshots of the carbon domains of all models at different stages of the thermals cycles. It reveals that, during the thermal annealing, carbon atoms undergo substantial organisation leading to the formation of dimers, chains and sheets. Specifically, we define five categories of carbon ...

    work page

  8. [8]

    Nucleation and growth mechanisms Figure 8a-f show local atomic configurations in the SiCNH8000-Random model at different times during the T = 1800 K plateau, highlighting the formation of an aromatic C ring. Starting from single carbon atoms and carbon dimers initially dispersed in the amorphous ceramic network (Figure 8a), we observe the progressive form...

    work page

  9. [9]

    Colombo, G

    P. Colombo, G. Mera, R. Riedel, and G. D. Sorarù, J. Am. Ceram. Soc.93, 1805 (2010)

    work page 2010

  10. [10]

    Riedel, Z

    R. Riedel, Z. Yu, S. Bernard, R. Kumar, and G. Singh, J. Am. Ceram. Soc.108, https://doi.org/10.1111/jace.70187 (2025)

    work page doi:10.1111/jace.70187 2025

  11. [11]

    Fritz and B

    G. Fritz and B. Raabe, Z. Anorg. Allg. Chem.286, 149 (1956)

    work page 1956

  12. [12]

    F. M. Ainger and J. M. Herbert,Special Ceramics, edited by P. Popper (Academic press, New York, US, 1960) pp. 168–182

    work page 1960

  13. [13]

    P. G. Chantrell and P. Popper,Special Ceramics, edited by P. Popper (Academic press, New York, US, 1965) pp. 87–103

    work page 1965

  14. [14]

    Colombo, G

    P. Colombo, G. Mera, R. Riedel, and G. D. Sorarù, J. Am. Ceram. Soc.93, 1805 (2010). 20

    work page 2010

  15. [15]

    Riedel, G

    R. Riedel, G. Mera, R. Hauser, and A. Klonczynski, J. Ceram. Soc. Jpn.114, 425 (2006)

    work page 2006

  16. [16]

    R. P. Chaudhary, C. Parameswaran, M. Idrees, A. S. Rasaki, C. Liu, Z. Chen, and P. Colombo, Prog. Mater. Sci.128, 100969 (2022)

    work page 2022

  17. [17]

    Cheype, V

    M. Cheype, V. Pateloup, and S. Bernard, Adv. Mater.36, 2307554 (2024)

    work page 2024

  18. [18]

    Z. C. Eckel, C. Zhou, J. H. Martin, A. J. Jacobsen, W. B. Carter, and T. A. Schaedler, Science351, 58 (2016)

    work page 2016

  19. [19]

    Barroso, Q

    G. Barroso, Q. Li, R. K. Bordia, and G. Motz, J. Mater. Chem. A7, 1936 (2019)

    work page 1936

  20. [20]

    Q. Wen, Z. Yu, and R. Riedel, Prog. Mater. Sci.109, 100623 (2020)

    work page 2020

  21. [21]

    Trassl, H.-J

    S. Trassl, H.-J. Kleebe, H. Störmer, G. Motz, E. Rössler, and G. Ziegler, J. Am. Ceram. Soc.85, 1268 (2002)

    work page 2002

  22. [22]

    Viard, D

    A. Viard, D. Fonblanc, M. Schmidt, A. Lale, C. Salameh, A. Soleilhavoup, M. Wynn, P. Champagne, S. Cerneaux, F. Babonneau,et al., Chem. Eur. J.23, 9076 (2017)

    work page 2017

  23. [23]

    Y. Chen, X. Yang, Y. Cao, Z. Gan, and L. An, Acta Mater.72, 22 (2014)

    work page 2014

  24. [24]

    Trassl, M

    S. Trassl, M. Puchinger, E. Rössler, and G. Ziegler, J. Eur. Ceram. Soc.23, 781 (2003)

    work page 2003

  25. [25]

    O. D. Adigun, E. Ricohermoso, A. A. Daniyan, L. E. Umoru, and E. Ionescu, Ceramics5, 690 (2022)

    work page 2022

  26. [26]

    Kleebe, H

    H.-J. Kleebe, H. Störmer, S. Trassl, and G. Ziegler, Appl. Organomet. Chem.15, 858 (2001)

    work page 2001

  27. [27]

    G. Mera, R. Riedel, F. Poli, and K. Müller, J. Eur. Ceram. Soc.29, 2873 (2009)

    work page 2009

  28. [28]

    Klausmann, K

    A. Klausmann, K. Morita, K. E. Johanns, C. Fasel, K. Durst, G. Mera, R. Riedel, and E. Ionescu, J. Eur. Ceram. Soc. 35, 3771 (2015)

    work page 2015

  29. [29]

    Ding and W

    J. Ding and W. Li, inAdvanced Ceramic Materials-Emerging Technologies(IntechOpen, 2024)

    work page 2024

  30. [30]

    R. K. Morais Ferreira, M. Ben Miled, R. K. Nishihora, N. Christophe, P. Carles, G. Motz, A. Bouzid, R. Machado, O. Masson, Y. Iwamoto, S. Célérier, A. Habrioux, and S. Bernard, Nanoscale Adv.5, 701 (2023)

    work page 2023

  31. [31]

    Ben Miled, M

    M. Ben Miled, M. Fradin, N. Benbakoura, L. Mazière, J. Rousseau, A. Bouzid, P. Carles, Y. Iwamoto, O. Masson, A. Habrioux, and S. Bernard, ChemSusChemn/a, e202400561 (2024)

    work page 2024

  32. [32]

    Widgeon, G

    S. Widgeon, G. Mera, Y. Gao, E. Stoyanov, S. Sen, A. Navrotsky, and R. Riedel, Chem. Mater.24, 1181 (2012)

    work page 2012

  33. [33]

    J. Dürr, P. Lamparter, J. Bill, S. Steeb, and F. Aldinger, J. Non-Cryst. Solids232-234, 155 (1998)

    work page 1998

  34. [34]

    Matsunaga, Y

    K. Matsunaga, Y. Iwamoto, C. A. J. Fisher, and H. Matsubara, J. Ceram. Soc. Jpn.107, 1025 (1999)

    work page 1999

  35. [35]

    N. Liao, W. Xue, and M. Zhang, J. Eur. Ceram. Soc.32, 1275 (2012)

    work page 2012

  36. [36]

    Kroll, J

    P. Kroll, J. Eur. Ceram. Soc.25, 163 (2005)

    work page 2005

  37. [37]

    Amkreutz and T

    M. Amkreutz and T. Frauenheim, Phys. Rev. B65, 134113 (2002)

    work page 2002

  38. [38]

    Mortier, S

    F. Mortier, S. Cadars, M. B. Miled., O. Masson, G. Ori, M. Boero, Y. Wang, S. Bernard, and A. Bouzid, Phys. Chem. Chem. Phys. 10.1039/D5CP04954G (2026)

    work page doi:10.1039/d5cp04954g 2026

  39. [39]

    Cadars, Python atomic modeling analyzes (2023)

    S. Cadars, Python atomic modeling analyzes (2023)

    work page 2023

  40. [40]

    Falgoust and P

    M. Falgoust and P. Kroll, J. Am. Ceram. Soc.107, 7653 (2024)

    work page 2024

  41. [41]

    Haseen and P

    S. Haseen and P. Kroll, J. Am. Ceram. Soc.109, e70392 (2026)

    work page 2026

  42. [42]

    Batatia, D

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csányi, Advances in neural information processing systems35, 11423 (2022)

    work page 2022

  43. [43]

    Batatia, S

    I. Batatia, S. Batzner, D. P. Kovács, A. Musaelian, G. N. C. Simm, R. Drautz, C. Ortner, B. Kozinsky, and G. Csányi, Nat. Mach. Intell.7, 56 (2025)

    work page 2025

  44. [44]

    T. D. Kühne, M. Iannuzzi, M. Del Ben, V. V. Rybkin, P. Seewald, F. Stein, T. Laino, R. Z. Khaliullin, O. Schütt, F. Schiffmann, D. Golze, J. Wilhelm, S. Chulkov, M. H. Bani-Hashemian, V. Weber, U. Borštnik, M. Taillefumier, A. S. Jakobovits, A. Lazzaro, H. Pabst, T. Müller, R. Schade, M. Guidon, S. Andermatt, N. Holmberg, G. K. Schenter, A. Hehn, A. Bussy...

    work page 2020

  45. [46]

    Goedecker, M

    S. Goedecker, M. Teter, and J. Hutter, Phys. Rev. B54, 1703 (1996)

    work page 1996

  46. [47]

    VandeVondele, M

    J. VandeVondele, M. Krack, F. Mohamed, M. Parrinello, T. Chassaing, and J. Hutter, Comput. Phys. Commun.167, 103 (2005)

    work page 2005

  47. [48]

    Grimme, J

    S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys.132, 154104 (2010)

    work page 2010

  48. [49]

    V. L. Deringer and G. Cs’anyi, Phys. Rev. B95, 094203 (2017)

    work page 2017

  49. [50]

    V. L. Deringer, N. Bernstein, A. P. Bartók, M. J. Cliffe, R. N. Kerber, L. E. Marbella, C. P. Grey, S. R. Elliott, and G. Csányi, J. Phys. Chem. Lett.9, 2879 (2018)

    work page 2018

  50. [51]

    A. R. Oganov and C. W. Glass, J. Chem. Phys.124, 244704 (2006)

    work page 2006

  51. [52]

    A. R. Oganov, A. O. Lyakhov, and M. Valle, Acc. Chem. Res.44, 227 (2011)

    work page 2011

  52. [53]

    A. O. Lyakhov, A. R. Oganov, H. T. Stokes, and Q. Zhu, Comput. Phys. Commun.184, 1172 (2013)

    work page 2013

  53. [54]

    Kresse and J

    G. Kresse and J. Hafner, Phys. Rev. B47, 558 (1993)

    work page 1993

  54. [55]

    Kresse and J

    G. Kresse and J. Furthm"uller, Phys. Rev. B54, 11169 (1996)

    work page 1996

  55. [56]

    J. B. Kruskal, Psychometrika29, 115 (1964)

    work page 1964

  56. [57]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg,et al., J. Mach. Learn. Res.12, 2825 (2011)

    work page 2011

  57. [58]

    Valle and A

    M. Valle and A. R. Oganov, Acta Crystallogr. Sect. A66, 507 (2010)

    work page 2010

  58. [59]

    D. P. Kov’acs, I. Batatia, E. S. Arany, and G. Cs’anyi, J. Chem. Phys.159, https://doi.org/10.1063/5.0155322 (2023)

    work page doi:10.1063/5.0155322 2023

  59. [60]

    Barrett, C

    R. Barrett, C. Ortner, and J. Westermayr, arXiv preprint arXiv:2502.12870 https://doi.org/10.48550/arXiv.2502.12870 (2025)

    work page doi:10.48550/arxiv.2502.12870 2025

  60. [61]

    A. Roy, R. Devanathan, S. Allec, G. Nandipati, A. M. Casella, D. J. Senor, D. D. Johnson, G. Balasubramanian, and A. Soulami, Adv. Intell. Discov. , e202500196 (2026). 21

    work page 2026

  61. [62]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization (2017)

    work page 2017

  62. [63]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton, Comput. Phys. Commun. 271, 108171 (2022)

    work page 2022

  63. [64]

    H. C. Edwards, C. R. Trott, and D. Sunderland, J. Parallel Distrib. Comput.74, 3202 (2014)

    work page 2014

  64. [65]

    Nosé, Mol

    S. Nosé, Mol. Phys.52, 255 (1984)

    work page 1984

  65. [66]

    S. Nosé, J. Chem. Phys.81, 511 (1984)

    work page 1984

  66. [67]

    W. G. Hoover, Phys. Rev. A31, 1695 (1985)

    work page 1985

  67. [68]

    J. M. Martínez and L. Martínez, J. Comput. Chem.24, 819 (2003)

    work page 2003

  68. [69]

    Martínez, R

    L. Martínez, R. Andrade, E. G. Birgin, and J. M. Martínez, J. Comput. Chem.30, 2157 (2009)

    work page 2009

  69. [71]

    Schempp, J

    S. Schempp, J. Dürr, P. Lamparter, J. Bill, and F. Aldinger, Zeitschrift für Naturforschung A53, 127 (1998)

    work page 1998

  70. [72]

    Choong Kwet Yive, R

    N. Choong Kwet Yive, R. Corriu, D. Leclercq, P. Mutin, and A. Vioux, Chem. Mater.4, 1263 (1992)

    work page 1992

  71. [73]

    Humphrey, A

    W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graph.14, 33 (1996)

    work page 1996

  72. [74]

    Seitz, J

    J. Seitz, J. Bill, N. Egger, and F. Aldinger, J. Eur. Ceram. Soc.16, 885 (1996)

    work page 1996

  73. [75]

    Boero, F

    M. Boero, F. Imoto, and A. Oshiyama, Phys. Rev. Mater.6, 093403 (2022)

    work page 2022

  74. [76]

    G. J. Martyna, M. L. Klein, and M. Tuckerman, J. Chem. Phys.97, 2635 (1992)

    work page 1992

  75. [77]

    T. D. Kühne, M. Krack, F. R. Mohamed, and M. Parrinello, Phys. Rev. Lett.98, 066401 (2007)

    work page 2007

  76. [78]

    T. D. Kühne, WIREs Computational Molecular Science4, 391 (2014)

    work page 2014

  77. [79]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett.77, 3865 (1996)

    work page 1996

  78. [80]

    Grimme, J

    S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys.132, 154104 (2010). 1 Supplementary Materials: Modeling phase separation in polymer-derived carbonitride ceramics through extended machine learning molecular dynamics Fabien Mortier, Sylvian Cadars, Olivier Masson, Mauro Boero, Guido Ori, Yun Wang, Samuel Bernard, and Assil Bouzid. S1. ADDITION...

    work page 2010