arxiv: 2604.09084 · v1 · submitted 2026-04-10 · ❄️ cond-mat.mtrl-sci

Recognition: unknown

Force Field-Agnostic Phase Classification of Zeolitic Imidazolate Framework Polymorphs

Emilio M\'endez (1) , L\'ena Triestram (2) , Dune Andr\'e (2) , Fran\c{c}ois-Xavier Coudert (2) , Rocio Semino (1) ((1) Sorbonne Universit\'e , CNRS , Physico-chimie des Electrolytes et Nanosyst\`emes Interfaciaux , PHENIX

show 6 more authors

Paris France (2) Chimie ParisTech PSL University Institut de Recherche de Chimie Paris France)

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:13 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci

keywords Zeolitic Imidazolate FrameworksPhase ClassificationNeural NetworksMolecular DynamicsForce FieldsPolymorphsZIF-4Metal-Organic Frameworks

0 comments

The pith

Neural network classifiers trained on multiple force fields distinguish ZIF polymorph phases without model-specific bias.

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

The paper shows that neural networks can classify phases of zeolitic imidazolate framework polymorphs directly from molecular dynamics trajectories. Classifiers reach high accuracy even with low-dimensional local structural descriptors. Training on configurations generated by several different force fields removes dependence on any single model and improves performance across simulations. The approach is then used to track the atomic-level steps of the ZIF-4-cp to ZIF-4-cp-II transition.

Core claim

Neural network classifiers built on local structural descriptors extracted from MD trajectories can accurately assign ZIF phases on the fly. When the training set combines data from multiple force fields, the resulting models become agnostic to the choice of force field, yielding higher accuracy and broader applicability than single-force-field training. Low-dimensional descriptors already suffice for reliable classification, while higher-dimensional inputs further improve results. Application to the ZIF-4-cp to ZIF-4-cp-II transition demonstrates the method's ability to reveal mechanistic details of the structural change.

What carries the argument

Neural network classifiers using local structural descriptors from MD configurations, trained across several force fields to achieve force-field-agnostic phase assignment.

If this is right

On-the-fly phase assignment becomes possible inside any MD run of ZIF materials without manual inspection.
Training on multiple force fields increases both accuracy and transferability of the classifier.
Low-dimensional descriptors already deliver high classification performance for these polymorphs.
Mechanistic pathways of phase transitions can be extracted automatically from simulation data.
The same workflow applies to other pairs of structurally close ZIF polymorphs.

Where Pith is reading between the lines

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

The method could be tested on additional ZIF systems or other metal-organic frameworks that exhibit polymorphism.
Combining the classifier with enhanced sampling techniques might accelerate discovery of new transition pathways.
If the descriptors prove transferable, the same networks could serve as order parameters in coarse-grained models of framework materials.

Load-bearing premise

Local structural descriptors taken from molecular-dynamics runs are sufficient to separate structurally similar ZIF phases consistently, and the network does not overfit to artifacts of any particular simulation model.

What would settle it

A new force field not included in training produces trajectories in which the classifier assigns phases with accuracy significantly below the reported levels or fails to identify the expected sequence of local changes during the ZIF-4-cp to ZIF-4-cp-II transition.

Figures

Figures reproduced from arXiv: 2604.09084 by (2) Chimie ParisTech, CNRS, Dune Andr\'e (2), Emilio M\'endez (1), Fran\c{c}ois-Xavier Coudert (2), France, France), Institut de Recherche de Chimie Paris, L\'ena Triestram (2), Paris, PHENIX, Physico-chimie des Electrolytes et Nanosyst\`emes Interfaciaux, PSL University, Rocio Semino (1) ((1) Sorbonne Universit\'e.

**Figure 2.** Figure 2: Workflow of the training and classification procedure for both classifiers. Trajec [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Figure 3: Confusion matrices of the SOAP- (left) and the BPSF-based (right) classifiers on [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

**Figure 4.** Figure 4: t-SNE embedding of the Zn-centred SOAP descriptors for all studied phases, from [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗

**Figure 5.** Figure 5: Fraction of Zn ions classified as CPII by the SOAP-based algorithm (left axis, full [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗

**Figure 6.** Figure 6: Snapshots of the phase transition from CP to CPII. Only Zn ions are represented [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗

**Figure 7.** Figure 7: Average cluster size ⟨li⟩ for directions i = x (black), y (red) and z (green), plotted against the number N of Zn ions comprising the cluster. Upper panel: CP to CPII transitions. Lower panel: CPII to CP transitions. The atoms that form the clusters correspond to the ones classified as belonging to the final phase in each case. Results from three independent simulations were averaged. 17 [PITH_FULL_IMAGE:… view at source ↗

**Figure 8.** Figure 8: Histograms of the angle ϕa between the vector normal to the surface formed by the four-membered rings of Zn atoms and the vector ⃗a that corresponds to the X axis of the simulation box. A sketch of the angle ϕa is depicted in the upper right corner, where each ball corresponds to a Zn atom forming a 4-membered ring, the vertical stick indicates the normal vector to the surface and the arrowed stick depicts… view at source ↗

read the original abstract

Zeolitic Imidazolate Frameworks (ZIFs) are a family of metal--organic frameworks that feature metal centers tetrahedrally linked to imidazole-based ligands and adopt zeolite-like topologies. ZIFs formed by Zinc cations and imidazolate linkers exhibit a remarkable degree of polymorphism, which can be modulated by varying synthesis parameters or thermodynamic conditions (i.e., temperature and pressure). Computer simulations provide a unique way of studying these materials and their phase transitions from the microscopic standpoint, revealing their underlying molecular mechanisms. However, studying these mechanisms requires to be able to classify the phase of each molecular entity in an agnostic and automatic fashion, which is particularly challenging when the two phases involved are structurally very similar. In this work, we systematically study neural network classifiers to classify ZIF phases on-the-fly during molecular dynamics simulations. We test a variety of input features, differing both in the dimensionality and nature of the descriptors and in the kind of force field used for building the training/testing database. We reveal that even with low-dimensional descriptors the classification is highly accurate, while the use of high-dimensional descriptors leads to an even better performance. Training the classifier with configurations coming from different force fields we can remove force field bias and enhance the classifier performance and general applicability. Finally, we apply our classifiers to reveal mechanistic details of the ZIF-4-cp $\xrightarrow{}$ ZIF-4-cp-II phase transition.

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

Training NNs on MD data from several force fields produces a more general classifier for similar ZIF phases and lets them track the cp to cp-II transition.

read the letter

The main point is that they train neural network classifiers on configurations drawn from multiple force fields so the model stops latching onto force-field-specific artifacts and instead picks up phase differences that hold across models. They test this on ZIF polymorphs, compare descriptor sets of varying size, and then run the classifier on the ZIF-4-cp to ZIF-4-cp-II transition to pull out mechanistic details during the simulation run. Even low-dimensional local descriptors already give high accuracy, and mixing the force fields improves both performance and transferability. That cross-force-field step is the practical advance; it is a simple way to make the labeling step less dependent on the choice of potential. The application to the transition shows the classifier can be used for something concrete rather than just sitting as a standalone method. The logic is straightforward and avoids the usual circularity of training and testing on the same force field. The soft spot is that the abstract supplies no accuracy numbers, no confusion matrices for the two close phases, and no explicit cross-validation or out-of-distribution results on a held-out force field. Without those figures it is hard to judge how large the reported gains actually are or whether the network is truly generalizing. The claim that local descriptors suffice across force fields is plausible but still needs the quantitative checks. This paper is for people who run molecular dynamics on ZIFs or related frameworks and need reliable on-the-fly phase assignment. A reader already working in that corner would find the training recipe and the transition analysis useful. I would send it to peer review; the idea is grounded enough and the target application is relevant, so referees can verify the numbers and tighten the validation.

Referee Report

2 major / 1 minor

Summary. The paper develops neural network classifiers for on-the-fly identification of ZIF polymorphs in MD simulations, focusing on structurally similar phases such as ZIF-4-cp and ZIF-4-cp-II. It systematically varies descriptor dimensionality and type, compares training sets derived from different force fields, and shows that multi-force-field training reduces bias while improving accuracy and transferability. The classifiers are then applied to extract mechanistic insights into the ZIF-4-cp to ZIF-4-cp-II phase transition.

Significance. If the quantitative results hold, the work supplies a robust, force-field-agnostic tool for phase classification in polymorphic MOFs, directly enabling mechanistic analysis of transitions that are otherwise difficult to resolve. The multi-force-field training strategy is a clear strength that addresses a common limitation in simulation-based studies. This could have practical value for the ZIF and broader MOF simulation community.

major comments (2)

[Results (classifier performance and multi-FF training)] The central claim that multi-force-field training removes bias and enhances generalizability is load-bearing, yet the provided abstract and summary give no numerical accuracy values, cross-validation statistics, or confusion-matrix details for the similar cp/cp-II phases. Without these, the magnitude and statistical significance of the reported improvement cannot be assessed.
[Methods and validation] The assumption that local structural descriptors extracted from MD trajectories suffice to distinguish phases across force fields without overfitting to simulation artifacts requires explicit out-of-distribution testing on held-out force fields; this is not quantified in the available text.

minor comments (1)

[Abstract] The abstract states that classification is 'highly accurate' with low-dimensional descriptors and 'even better' with high-dimensional ones, but supplies no concrete accuracy, precision, or recall figures. Adding these numbers would strengthen the summary.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and the recommendation of minor revision. The comments are constructive and help strengthen the presentation of our results on force-field-agnostic classification. Below we respond point by point to the major comments.

read point-by-point responses

Referee: [Results (classifier performance and multi-FF training)] The central claim that multi-force-field training removes bias and enhances generalizability is load-bearing, yet the provided abstract and summary give no numerical accuracy values, cross-validation statistics, or confusion-matrix details for the similar cp/cp-II phases. Without these, the magnitude and statistical significance of the reported improvement cannot be assessed.

Authors: We agree that quantitative metrics should be more immediately accessible. The full manuscript reports these values in the Results section (accuracy, 5-fold cross-validation scores, and confusion matrices for the cp/cp-II pair under single- versus multi-force-field training). The multi-force-field models show improved accuracy and lower misclassification rates between the structurally similar phases, with statistical significance evaluated via appropriate tests. To address the referee's concern directly, we will revise the abstract to include representative numerical values and add a compact summary table of key performance metrics. revision: partial
Referee: [Methods and validation] The assumption that local structural descriptors extracted from MD trajectories suffice to distinguish phases across force fields without overfitting to simulation artifacts requires explicit out-of-distribution testing on held-out force fields; this is not quantified in the available text.

Authors: This is a fair point. Our multi-force-field training protocol already incorporates training on data from several force fields while evaluating on configurations generated by the remaining ones, which provides a measure of cross-force-field transferability. Nevertheless, we acknowledge that a more explicit held-out-force-field protocol would strengthen the validation. In the revised manuscript we will add a dedicated paragraph and supporting figure that quantifies classifier performance when one force field is entirely excluded from training and used solely for testing, thereby confirming that the descriptors do not overfit to force-field-specific artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper trains neural network classifiers on local structural descriptors from MD trajectories of known ZIF phases simulated with multiple force fields. Training data labels derive from simulation conditions and phase identities, not from the classifier outputs. Evaluation uses held-out configurations and cross-force-field tests, supplying independent performance metrics. No derivation step, equation, or claim reduces by construction to its own inputs or to a self-citation chain; the central result (bias removal via multi-FF training and application to the cp to cp-II transition) rests on empirical generalization rather than self-definition or fitted renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on standard assumptions of molecular dynamics and supervised learning rather than new physical postulates.

axioms (1)

domain assumption Local atomic descriptors computed from MD snapshots contain sufficient information to distinguish ZIF polymorphs even when the phases are structurally similar.
Invoked by the choice of input features for the neural network classifiers.

pith-pipeline@v0.9.0 · 5649 in / 1207 out tokens · 79642 ms · 2026-05-10T17:13:12.845773+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

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

Machine Learning and Molecular Simulations Reveal Mechanisms of ZIFs Polymorph Selection
cond-mat.mtrl-sci 2026-04 unverdicted novelty 6.0

Polymorph selection in ZIFs occurs at the pre-nucleation cluster stage, as revealed by metadynamics simulations and neural network classification of intermediate structures.

Reference graph

Works this paper leans on

41 extracted references · 5 canonical work pages · cited by 1 Pith paper · 2 internal anchors

[1]

Schneemann, A.; Bon, V.; Schwedler, I.; Senkovska, I.; Kaskel, S.; Fischer, R. A. Flexible metal–organic frameworks. Chem. Soc. Rev. 2014, 43, 6062–6096

2014
[2]

B.; Karmakar, A.; Zhao, D

Peh, S. B.; Karmakar, A.; Zhao, D. Multiscale Design of Flexible Metal–Organic Frameworks. Trends Chem. 2020, 2, 199–213

2020
[3]

Recent advances in stimuli-responsive framework materials: Understanding their response and searching for materials with targeted behavior

Coudert, F.-X. Recent advances in stimuli-responsive framework materials: Understanding their response and searching for materials with targeted behavior. Coord. Chem. Rev. 2025, 539, 216760

2025
[4]

R.; Wriedt, M

Aulakh, D.; Varghese, J. R.; Wriedt, M. The Importance of Polymorphism in Metal–Organic Framework Studies. Inorg. Chem. 2015, 54, 8679–8684

2015
[5]

K.; Kieslich, G.; Yeung, H

Cheetham, A. K.; Kieslich, G.; Yeung, H. H.-M. Thermodynamic and Kinetic Effects in the Crystallization of Metal–Organic Frameworks. Acc. Chem. Res. 2018, 51, 659–667

2018
[6]

M.; Keen, D

Castillo-Blas, C.; Chester, A. M.; Keen, D. A.; Bennett, T. D. Thermally activated structural phase transitions and processes in metal–organic frameworks. Chem. Soc. Rev. 2024, 53, 3606–3629

2024
[7]

T.; Bennett, T

Hughes, J. T.; Bennett, T. D.; Cheetham, A. K.; Navrotsky, A. Thermochemistry of Zeolitic Imidazolate Frameworks of Varying Porosity. J. Am. Chem. Soc. 2012, 135, 598–601

2012
[8]

N.; Lampronti, G

Widmer, R. N.; Lampronti, G. I.; Chibani, S.; Wilson, C. W.; Anzellini, S.; Farsang, S.; Kleppe, A. K.; Casati, N. P. M.; MacLeod, S. G.; Redfern, S. A. T.; Coudert, F.-X.; Bennett, T. D. Rich Polymorphism of a Metal--Organic Framework in Pressure--Temperature Space. J. Am. Chem. Soc. 2019, 141, 9330--9337

2019
[9]

M.; Ríos Gómez, M

Bumstead, A. M.; Ríos Gómez, M. L.; Thorne, M. F.; Sapnik, A. F.; Longley, L.; Tuffnell, J. M.; Keeble, D. S.; Keen, D. A.; Bennett, T. D. Investigating the melting behaviour of polymorphic zeolitic imidazolate frameworks. CrystEngComm 2020, 22, 3627–3637

2020
[10]

F.; Bennett, T

Hou, J.; Sapnik, A. F.; Bennett, T. D. Metal–organic framework gels and monoliths. Chem. Sci. 2020, 11, 310–323

2020
[11]

R.; Wondraczek, L.; K\" a rger, J.; Knebel, A

Smirnova, O.; Hwang, S.; Sajzew, R.; Ge, L.; Reupert, A.; Nozari, V.; Savani, S.; Chmelik, C.; Reithofer, M. R.; Wondraczek, L.; K\" a rger, J.; Knebel, A. Precise control over gas-transporting channels in zeolitic imidazolate framework glasses. Nature Mater. 2023, 23, 262–270

2023
[12]

Phase diagram of ZIF-4 from computer simulations

Méndez, E.; Semino, R. Phase diagram of ZIF-4 from computer simulations. J. Mater. Chem. A 2024, 12, 31108--31115

2024
[13]

S.; Ni, Z.; C\^oté, A

Park, K. S.; Ni, Z.; C\^oté, A. P.; Choi, J. Y.; Huang, R.; Uribe-Romo, F. J.; Chae, H. K.; O’Keeffe, M.; Yaghi, O. M. Exceptional chemical and thermal stability of zeolitic imidazolate frameworks. Proc. Nat. Acad. Sci. 2006, 103, 10186–10191

2006
[14]

Vervoorts, P.; Stebani, J.; Méndez, A. S. J.; Kieslich, G. Structural Chemistry of Metal–Organic Frameworks under Hydrostatic Pressures. ACS Mater. Lett. 2021, 3, 1635–1651

2021
[15]

L.; Henke, S

Song, J.; Pallach, R.; Frentzel‐Beyme, L.; Kolodzeiski, P.; Kieslich, G.; Vervoorts, P.; Hobday, C. L.; Henke, S. Tuning the High‐Pressure Phase Behaviour of Highly Compressible Zeolitic Imidazolate Frameworks: From Discontinuous to Continuous Pore Closure by Linker Substitution. Angew. Chem. Int. Ed. 2022, 61, e202117565

2022
[16]

P.; Anzellini, S.; Meneghini, C.; Herlihy, A.; Amboage, M.; Keen, D

Robertson, G. P.; Anzellini, S.; Meneghini, C.; Herlihy, A.; Amboage, M.; Keen, D. A.; Irifune, T.; Bennett, T. D. Changes in the Long-Range Order and Local Atomic Structure of Zeolitic Imidazolate Frameworks under Extreme Conditions. Inorg. Chem. 2025, 65, 156–164

2025
[17]

U.; Boutin, A.; Coudert, F.-X

Bouëssel du Bourg, L.; Ortiz, A. U.; Boutin, A.; Coudert, F.-X. Thermal and mechanical stability of zeolitic imidazolate frameworks polymorphs. APL Mater. 2014, 2, 124110

2014
[18]

D.; Coudert, F.-X

Castel, N.; André, D.; Edwards, C.; Evans, J. D.; Coudert, F.-X. Machine learning interatomic potentials for amorphous zeolitic imidazolate frameworks. Digit. Discov. 2024, 3, 355--368

2024
[19]

Challenges in Molecular Dynamics of Amorphous ZIFs Using Reactive Force Fields

Castel, N.; Coudert, F.-X. Challenges in Molecular Dynamics of Amorphous ZIFs Using Reactive Force Fields . J. Phys. Chem. C 2022, 126, 19532--19541

2022
[20]

Microscopic mechanism of thermal amorphization of ZIF -4 and melting of ZIF -zni revealed via molecular dynamics and machine learning techniques

Méndez, E.; Semino, R. Microscopic mechanism of thermal amorphization of ZIF -4 and melting of ZIF -zni revealed via molecular dynamics and machine learning techniques. J. Mater. Chem. A 2024, 12, 4572--4582

2024
[21]

A.; Semino, R.; Pireddu, G.; Auerbach, S

Helfrecht, B. A.; Semino, R.; Pireddu, G.; Auerbach, S. M.; Ceriotti, M. A new kind of atlas of zeolite building blocks. J. Chem. Phys. 2019, 151, 154112

2019
[22]

M.; Schmidt, S.; Schiøtz, J

Larsen, P. M.; Schmidt, S.; Schiøtz, J. Robust structural identification via polyhedral template matching. Model. Simul. Mater. Sci. Eng. 2016, 24, 055007

2016
[23]

Rosset, L. A. M.; Drabold, D. A.; Deringer, V. L. Signatures of paracrystallinity in amorphous silicon from machine-learning-driven molecular dynamics. Nature Commun. 2025, 16, 2360

2025
[24]

Atom-centered symmetry functions for constructing high-dimensional neural network potentials

Behler, J. Atom-centered symmetry functions for constructing high-dimensional neural network potentials. J. Chem. Phys. 2011, 134, 074106

2011
[25]

P.; Kondor, R.; Csányi, G

Bartók, A. P.; Kondor, R.; Csányi, G. On representing chemical environments. Phys. Rev. B 2013, 87, 184115

2013
[26]

T.; Kieslich, G.; Hante, I.; Schneemann, A.; Wu, Y.; Daisenberger, D.; Cheetham, A

Henke, S.; Wharmby, M. T.; Kieslich, G.; Hante, I.; Schneemann, A.; Wu, Y.; Daisenberger, D.; Cheetham, A. K. Pore closure in zeolitic imidazolate frameworks under mechanical pressure. Chem. Sci. 2018, 9, 1654–1660

2018
[27]

Balestra, S. R. G.; Semino, R. Computer simulation of the early stages of self-assembly and thermal decomposition of ZIF-8. J. Chem. Phys. 2022, 157, 184502

2022
[28]

Identifying phase transitions in zeolitic imidazolate frameworks: microscopic insight from molecular simulations

Triestram, L.; Coudert, F.-X. Identifying phase transitions in zeolitic imidazolate frameworks: microscopic insight from molecular simulations. Chem. Sci. 2026, 17, 6734–6745

2026
[29]

Unveiling ZIF-8 Nucleation Mechanisms through Molecular Simulation: Role of Temperature, Solvent, and Reactant Concentration

Andarzi Gargari, S.; Semino, R. Unveiling ZIF-8 Nucleation Mechanisms through Molecular Simulation: Role of Temperature, Solvent, and Reactant Concentration. Chem. Mater. 2025, 37, 9460–9470

2025
[30]

Thermodynamic insights into the self-assembly of zeolitic imidazolate frameworks from computer simulations

Méndez, E.; Semino, R. Thermodynamic insights into the self-assembly of zeolitic imidazolate frameworks from computer simulations. Chem. Sci. 2025, 16, 11979–11988

2025
[31]

P.; Simm, G.; Ortner, C.; Cs \'a nyi, G

Batatia, I.; Kovacs, D. P.; Simm, G.; Ortner, C.; Cs \'a nyi, G. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields. Advances in neural information processing systems 2022, 35, 11423--11436

2022
[32]

Batatia, D

Batatia, I.; Kovács, D. P.; Simm, G. N. C.; Ortner, C.; Csányi, G. MACE : Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields. 2023; http://arxiv.org/abs/2206.07697, arXiv:2206.07697

work page arXiv 2023
[33]

Visualizing Data using t-SNE

van der Maaten, L.; Hinton, G. Visualizing Data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579--2605

2008
[34]

Tanaka, H.; Ohsaki, S.; Hiraide, S.; Yamamoto, D.; Watanabe, S.; Miyahara, M. T. Adsorption-Induced Structural Transition of ZIF-8: A Combined Experimental and Simulation Study. J. Phys. Chem. C 2014, 118, 8445–8454

2014
[35]

H.; Coudert, F.-X.; Patarin, J

Chaplais, G.; Fraux, G.; Paillaud, J.-L.; Marichal, C.; Nouali, H.; Fuchs, A. H.; Coudert, F.-X.; Patarin, J. Impacts of the Imidazolate Linker Substitution (CH 3 , Cl, or Br) on the Structural and Adsorptive Properties of ZIF-8. J. Phys. Chem. C 2018, 122, 26945–26955

2018
[36]

J.; Kohlmeyer, A.; Thompson, A

Plimpton, S. J.; Kohlmeyer, A.; Thompson, A. P.; Moore, S. G.; Berger, R. LAMMPS : Large -scale Atomic / Molecular Massively Parallel Simulator . 2023; https://zenodo.org/records/10806852

work page arXiv 2023
[37]

R.; McCarthy, M

Johansson, A.; Weinberg, E.; Trott, C. R.; McCarthy, M. J.; Moore, S. G. LAMMPS-KOKKOS: Performance Portable Molecular Dynamics Across Exascale Architectures. 2025; https://arxiv.org/abs/2508.13523

work page arXiv 2025
[38]

Trott, C. R. et al. Kokkos 3: Programming Model Extensions for the Exascale Era. IEEE Transactions on Parallel and Distributed Systems 2022, 33, 805--817

2022
[39]

Pedregosa, F. et al. Scikit-learn: Machine Learning in P ython. J. Mach. Learn. Res. 2011, 12, 2825--2830

2011
[40]

Agarap, A. F. Deep Learning using Rectified Linear Units ( ReLU ). 2019; http://arxiv.org/abs/1803.08375, arXiv:1803.08375 [cs]

work page internal anchor Pith review Pith/arXiv arXiv 2019
[41]

Adam: A Method for Stochastic Optimization

Kingma, D. P.; Ba, J. Adam: A Method for Stochastic Optimization . 2017; http://arxiv.org/abs/1412.6980, arXiv:1412.6980 [cs] mcitethebibliography arxiv.tex0000664000000000000000000015435715166117510011453 0ustar rootroot [journal=jacsat,manuscript=article] achemso [version=3] mhchem subcaption xcolor hyperref Chimie ParisTech, PSL University, CNRS, Insti...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1021/jacs.9b03234 2017