A Review of Force Field Models Based on Machine Learning

doi:10.11871/jfdc.issn.2096-742X.2023.04.003

Abstract

Abstract:

[Background] In the past few decades, due to complexity of the atomic structure and the computation for investigating the structure, traditional force field methods have been struggling in solving certain problems. [Objective] The introduction of machine learning methods is expected to solve many previously intractable problems and balance the constraints between computational efficiency and accuracy. [Methods] Machine learning force fields methods do not rely on preconceived knowledge and model the force field by learning from small-scale high-precision molecular dynamics simulation data. At the same time, approximate assumptions are made for the motion of atomic nuclei and extranuclear electrons, greatly simplifying the generation process of the force field. [Results] Machine learning force fields methods aim to achieve almost the same accuracy as traditional force fields methods while significantly improving computational efficiency. This article provides an overview of the development and related theoretical knowledge of the machine learning force fields methods, introduces several common methods, and finally explores the shortcomings of machine learning force fields methods and the challenges that need to be tackled in the future.

Key words: machine learning in semiconductor field, machine learning force field, sGDML, neural network

CHEN Meilin, LIU Duanyang, XU Liming, WANG Yang. A Review of Force Field Models Based on Machine Learning[J]. Frontiers of Data and Computing, 2023, 5(4): 27-37, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2023.04.003.

Figures/Tables 3

References 51

[1]	LIU D, XU L, LIN X, et al. Machine Learning for Semi-conductors[J]. Chip, 2022, 1(4): 100033. doi: 10.1016/j.chip.2022.100033
[2]	FUKUHARA S, SHIMOJO F, SHIBUTA Y. Confor-mation and catalytic activity of nickel-carbon cluster for ethanol dissociation in carbon nanotube synthesis: Ab initio molecular dynamics simulation[J]. Chemical Physics Letters, 2017, 679: 164-171. doi: 10.1016/j.cplett.2017.04.086
[3]	CHMIELA S, SAUCEDA H E, MÜLLER K R, et al. Tow-ards exact molecular dynamics simulations with machine-learned force fields[J]. Nature communications, 2018, 9(1): 3887. doi: 10.1038/s41467-018-06169-2
[4]	TUCKERMAN M E. molecular dynamics: basic conc-epts, current trends and novel applications[J]. Journal of Physics Condensed Matter, 14(50): R1297-R1355. doi: 10.1088/0953-8984/14/50/202
[5]	CHU Z M, XIAO D J, QIAO Y S, et al. Machine Learn-ing Applications for Particle Accelerators[J]. Frontiers of Data and Computing, 2019, 1(2): 110-120.
[6]	UNKE O T, CHMIELA S, SAUCEDA H E, et al. Mach-ine learning force fields[J]. Chemical Reviews, 2021, 121(16): 10142-10186. doi: 10.1021/acs.chemrev.0c01111
[7]	BARTÓK A P, PAYNE M C, KONDOR R, et al. Gaus-sian approximation potentials: The accuracy of quantum mechanics, without the electrons[J]. Physical review letters, 2010, 104(13): 136403. doi: 10.1103/PhysRevLett.104.136403
[8]	UNKE O T, MEUWLY M. PhysNet: A neural network for predicting energies, forces, dipole moments, and partial charges[J]. Journal of chemical theory and compu-tation, 2019, 15(6): 3678-3693.
[9]	BEHLER J, PARRINELLO M. Generalized neural-network representation of high-dimensional potential-energy surfaces[J]. Physical review letters, 2007, 98(14): 146401. doi: 10.1103/PhysRevLett.98.146401
[10]	ZHANG L, HAN J, WANG H, et al. Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics[J]. Physical review letters, 2018, 120(14): 143001. doi: 10.1103/PhysRevLett.120.143001
[11]	STÖHR M, MEDRANO SANDONAS L, TKATC-HENKO A. Accurate many-body repulsive potentials for density-functional tight binding from deep tensor neural networks[J]. The Journal of Physical Chemistry Letters, 2020, 11(16): 6835-6843. doi: 10.1021/acs.jpclett.0c01307
[12]	CHMIELA S, VASSILEV-GALINDO V, UNKE O T, et al. Accurate global machine learning force fields for molecules with hundreds of atoms[J]. Science Advances, 2023, 9(2): eadf0873.
[13]	GUO J L, WANG Z G, WANG Y G, et al. A Review of Material Research and Development Methods Based on Computer Technology[J]. Frontiers of Data and Comp-uting, 2021, 3(2): 120-132.
[14]	WANG Z G, WAN M, CHEN Z Y, et al. Research and Application of a Data-Driven Intelligent Design Platform for Materials[J]. Frontiers of Data and Computing, 2023, 5(2): 86-96.
[15]	SZE V, CHEN Y H, YANG T J, et al. Efficient processing of deep neural networks: A tutorial and survey[J]. Proceedings of the IEEE, 2017, 105(12): 2295-2329. doi: 10.1109/JPROC.2017.2761740
[16]	BARTÓK A P, KERMODE J, BERNSTEIN N, et al. Machine learning a general-purpose interatomic potential for silicon[J]. Physical Review X, 2018, 8(4): 041048. doi: 10.1103/PhysRevX.8.041048
[17]	SAUCEDA H E, VASSILEV-GALINDO V, CHMIELA S, et al. Dynamical strengthening of covalent and non-covalent molecular interactions by nuclear quantum effe-cts at finite temperature[J]. Nature Communications, 2021, 12(1): 442. doi: 10.1038/s41467-020-20212-1
[18]	SCHÜTT K T, ARBABZADAH F, CHMIELA S, et al. Quantum-chemical insights from deep tensor neural networks[J]. Nature communications, 2017, 8(1): 13890. doi: 10.1038/ncomms13890
[19]	申林, 贾璐阳, 汤典东, 等. 计算化学中的机器学习[J]. 中国科学:化学, 2022, 52(6): 858-868
[20]	POLTAVSKY I, TKATCHENKO A. Machine learn-ing force fields: Recent advances and remaining challe-nges[J]. The Journal of Physical Chemistry Letters, 2021, 12(28): 6551-6564. doi: 10.1021/acs.jpclett.1c01204
[21]	WEI J, CHU X, SUN X Y, et al. Machine learning in materials science[J]. InfoMat, 2019, 1(3): 338-358. doi: 10.1002/inf2.v1.3
[22]	TU Y Y, ZHENG Q J, ZHAO J. Research on Quantum Proton Coupled Charge Transfer Process Based on Deep Neural Network[J]. Frontiers of Data and Computing, 2023, 5(2): 37-49.
[23]	ROSENBERGER D, SMITH J S, GARCIA A E. Mod-eling of peptides with classical and novel machine learning force fields: A comparison[J]. The Journal of Physical Chemistry B, 2021, 125(14): 3598-3612. doi: 10.1021/acs.jpcb.0c10401
[24]	SMITH J S, NEBGEN B T, ZUBATYUK R, et al. Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer lear-ning[J]. Nature communications, 2019, 10(1): 2903. doi: 10.1038/s41467-019-10827-4
[25]	SMITH J S, ISAYEV O, ROITBERG A E. ANI-1: an extensible neural network potential with DFT accuracy at force field computational cost[J]. Chemical science, 2017, 8(4): 3192-3203. doi: 10.1039/c6sc05720a pmid: 28507695
[26]	SMITH J S, ZUBATYUK R, NEBGEN B, et al. The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules[J]. Scientific data, 2020, 7(1): 134. doi: 10.1038/s41597-020-0473-z pmid: 32358545
[27]	HU W, SHUAIBI M, DAS A, et al. Forcenet: A graph neural network for large-scale quantum calculations[J]. arXiv preprint arXiv:2103.01436, 2021.
[28]	CHMIELA S, TKATCHENKO A, SAUCEDA H E, et al. Machine learning of accurate energy-conserving molecular force fields[J]. Science advances, 2017, 3(5): e1603015.
[29]	CHMIELA S, SAUCEDA H E, POLTAVSKY I, et al. sGDML: Constructing accurate and data efficient mole-cular force fields using machine learning[J]. Computer Physics Communications, 2019, 240: 38-45. doi: 10.1016/j.cpc.2019.02.007
[30]	GASTEIGER J, GROß J, GÜNNEMANN S. Directional message passing for molecular graphs[J]. arXiv preprint arXiv:2003.03123, 2020.
[31]	GASTEIGER J, GIRI S, MARGRAF J T, et al. Fast and uncertainty-aware directional message passing for non-equilibrium molecules[J]. arXiv preprint arXiv: 2011.14115, 2020.
[32]	SCHÜTT K T, HESSMANN S S P, GEBAUER N W A, et al. SchNetPack 2.0: A neural network toolbox for atomistic machine learning[J]. The Journal of Chemical Physics, 2023, 158(14): 144801. doi: 10.1063/5.0138367
[33]	SCHÜTT K, KINDERMANS P J, SAUCEDA FELIX H E, et al. Schnet: A continuous-filter convolutional neural network for modeling quantum interactions[J]. Advances in neural information processing systems, 2017, 30: 991-1001.
[34]	ZENI C, ROSSI K, GLIELMO A, et al. On machine learning force fields for metallic nanoparticles[J]. Adv-ances in Physics: X, 2019, 4(1): 1654919.
[35]	ZHANG W, WENG M, ZHANG M, et al. Revealing Morphology Evolution of Lithium Dendrites by Large-Scale Simulation Based on Machine Learning Force Field[J]. Advanced Energy Materials, 2023, 13(4): 2202892. doi: 10.1002/aenm.v13.4
[36]	LI Z, MEIDANI K, YADAV P, et al. Graph neural net-works accelerated molecular dynamics[J]. The Journal of Chemical Physics, 2022, 156(14): 144103. doi: 10.1063/5.0083060
[37]	KABYLDA A, VASSILEV-GALINDO V, CHMIELA S, et al. Towards linearly scaling and chemically accurate global machine learning force fields for large molecules[J]. arXiv preprint arXiv:2209.03985, 2022.
[38]	SAUCEDA H E, GÁLVEZ-GONZÁLEZ L E, CHM-IELA S, et al. BIGDML: Towards Exact Machine Lear-ning Force Fields for Materials[J]. arXiv e-prints, 2021: arXiv: 2106.04229.
[39]	GASTEGGER M, SCHWIEDRZIK L, BITTERMANN M, et al. wACSF—Weighted atom-centered symmetry functions as descriptors in machine learning potentials[J]. The Journal of chemical physics, 2018, 148(24): 241709. doi: 10.1063/1.5019667
[40]	LUBBERS N, SMITH J S, BARROS K. Hierarchical modeling of molecular energies using a deep neural network[J]. The Journal of chemical physics, 2018, 148(24): 241715. doi: 10.1063/1.5011181
[41]	SCHÜTT K T, KESSEL P, GASTEGGER M, et al. Sch-NetPack: A deep learning toolbox for atomistic syste-ms[J]. Journal of chemical theory and computation, 2018, 15(1): 448-455. doi: 10.1021/acs.jctc.8b00908
[42]	SCHÜTT K T, SAUCEDA H E, KINDERMANS P J, et al. Schnet-a deep learning architecture for molecules and materials[J]. The Journal of Chemical Physics, 2018, 148 (24): 241722. doi: 10.1063/1.5019779
[43]	BATTAGLIA P, PASCANU R, LAI M, et al. Interaction networks for learning about objects, relations and physics[J]. Advances in neural information processing systems, 2016, 29: 4509-4517.
[44]	XIE T, GROSSMAN J C. Crystal graph convolutional neural networks for an accurate and interpretable pre-diction of material properties[J]. Physical review letters, 2018, 120(14): 145301. doi: 10.1103/PhysRevLett.120.145301
[45]	FU X, WU Z, WANG W, et al. Forces are not enough: Benchmark and critical evaluation for machine learning force fields with molecular simulations[J]. arXiv preprint arXiv:2210.07237, 2022.
[46]	GKEKA P, STOLTZ G, BARATI FARIMANI A, et al. Machine learning force fields and coarse-grained variables in molecular dynamics: application to materials and biological systems[J]. Journal of chemical theory and computation, 2020, 16(8): 4757-4775. doi: 10.1021/acs.jctc.0c00355 pmid: 32559068
[47]	MISHIN Y. Machine-learning interatomic potentials for materials science[J]. Acta Materialia, 2021, 214: 116980. doi: 10.1016/j.actamat.2021.116980
[48]	SCHMIDT J, MARQUES M R G, BOTTI S, et al. Recent advances and applications of machine learning in solid-state materials science[J]. Npj Computational Materials, 2019, 5(1): 83. doi: 10.1038/s41524-019-0221-0
[49]	MILARDOVICH D, WALDHOER D, JECH M, et al. Building robust machine learning force fields by composite Gaussian approximation potentials[J]. Solid-State Electronics, 2023, 200: 108529. doi: 10.1016/j.sse.2022.108529
[50]	SCHMITZ N F, MÜLLER K R, CHMIELA S. Alg-orithmic differentiation for automated modeling of machine learned force fields[J]. The Journal of Physical Chemistry Letters, 2022, 13(43): 10183-10189. doi: 10.1021/acs.jpclett.2c02632
[51]	LIAO K, DONG S, CHENG Z, et al. Combined frag-ment-based machine learning force field with classical force field and its application in the NMR calculations of macromolecules in solutions[J]. Physical Chemistry Chemical Physics, 2022, 24(31): 18559-18567. doi: 10.1039/D2CP02192G

	GDML		sGDML
	Force	Energy	Force	Energy
Benzene	0.195	0.07	0.16	0.07
Uracil	0.663	0.142	0.663	0.142
Naphthalene	0.222	0.12	0.113	0.116
Aspirin	0.984	0.264	0.679	0.194
Salicylic Acid	0.829	0.178	0.829	0.178
Malonaldehyde	0796	0.157	0.414	0.098
Ethanol	0.792	0.154	0.335	0.072
Toluene	0.425	0.125	0.142	0.097
Paracetamol	1.036	0.274	0.491	0.153
Azobenzene	0.78	0.353	0.409	0.092

	SchNet		sGDML
	Force	Energy	Force	Energy
Benzene	0.31	0.08	0.16	0.07
Uracil	0.56	0.17	0.663	0.142
Naphthalene	0.58	0.16	0.113	0.116
Aspirin	0.135	0.37	0.679	0.194
Salicylic acid	0.85	0.2	0.829	0.178
Malonaldehyde	0.66	0.13	0.414	0.098
Ethanol	0.39	0.08	0.335	0.072
Toluene	0.57	0.12	0.142	0.097