Crossref journal-article
AIP Publishing
The Journal of Chemical Physics (317)
Abstract

Despite the ever-increasing computer power, accurate ab initio calculations for large systems (thousands to millions of atoms) remain infeasible. Instead, approximate empirical energy functions are used. Most current approaches are either transferable between different chemical systems, but not particularly accurate, or they are fine-tuned to a specific application. In this work, a data-driven method to construct a potential energy surface based on neural networks is presented. Since the total energy is decomposed into local atomic contributions, the evaluation is easily parallelizable and scales linearly with system size. With prediction errors below 0.5 kcal mol−1 for both unknown molecules and configurations, the method is accurate across chemical and configurational space, which is demonstrated by applying it to datasets from nonreactive and reactive molecular dynamics simulations and a diverse database of equilibrium structures. The possibility to use small molecules as reference data to predict larger structures is also explored. Since the descriptor only uses local information, high-level ab initio methods, which are computationally too expensive for large molecules, become feasible for generating the necessary reference data used to train the neural network.

Bibliography

Unke, O. T., & Meuwly, M. (2018). A reactive, scalable, and transferable model for molecular energies from a neural network approach based on local information. The Journal of Chemical Physics, 148(24).

Authors 2
  1. Oliver T. Unke (first)
  2. Markus Meuwly (additional)
References 121 Referenced 79
  1. 10.1098/rspa.1929.0094 / Proc. R. Soc. A (1929)
  2. 10.1007/978-94-009-9027-2_2 / Horizons of Quantum Chemistry (1980)
  3. 10.1002/(sici)1521-3773(19990712)38:13/14<1894::aid-anie1894>3.0.co;2-h / Angew. Chem., Int. Ed. (1999)
  4. 10.1002/nme.1620151110 / Int. J. Numer. Methods Eng. (1980)
  5. 10.1063/1.470536 / J. Chem. Phys. (1995)
  6. 10.1063/1.479368 / J. Chem. Phys. (1999)
  7. 10.2307/2007507 / Math. Comput. (1981)
  8. 10.1063/1.466801 / J. Chem. Phys. (1994)
  9. 10.1063/1.2831790 / J. Chem. Phys. (2008)
  10. 10.1007/s10910-008-9354-y / J. Math. Chem. (2008)
  11. 10.1080/01442350903234923 / Int. Rev. Phys. Chem. (2009)
  12. 10.1063/1.4811653 / J. Chem. Phys. (2013)
  13. 10.1063/1.470984 / J. Chem. Phys. (1996)
  14. 10.1063/1.473683 / J. Chem. Phys. (1997)
  15. 10.1146/annurev.physchem.50.1.537 / Annu. Rev. Phys. Chem. (1999)
  16. 10.1021/acs.jcim.7b00090 / J. Chem. Inf. Model. (2017)
  17. 10.1021/acs.jcim.6b00280 / J. Chem. Inf. Model. (2016)
  18. 10.1002/anie.201403924 / Angew. Chem., Int. Ed. (2014)
  19. 10.1002/anie.201403689 / Angew. Chem., Int. Ed. (2014)
  20. 10.1021/acs.jpcb.7b03570 / J. Phys. Chem. B (2017)
  21. 10.1002/jcc.540040211 / J. Comput. Chem. (1983)
  22. 10.1021/ja00205a001 / J. Am. Chem. Soc. (1989)
  23. 10.1021/j100389a010 / J. Phys. Chem. (1990)
  24. 10.1021/ja00051a040 / J. Am. Chem. Soc. (1992)
  25. 10.1021/ja00124a002 / J. Am. Chem. Soc. (1995)
  26. 10.1002/jcc.20082 / J. Comput. Chem. (2004)
  27. 10.1021/ja00063a043 / J. Am. Chem. Soc. (1993)
  28. 10.1039/b604290b / Chem. Commun. (2006)
  29. 10.1021/ct700134r / J. Chem. Theory Comput. (2007)
  30. 10.1021/ct800392n / J. Chem. Theory Comput. (2009)
  31. 10.1103/physrevb.46.2727 / Phys. Rev. B (1992)
  32. 10.1088/0965-0393/2/1/011 / Modell. Simul. Mater. Sci. Eng. (1994)
  33. 10.1103/physrevlett.56.632 / Phys. Rev. Lett. (1986)
  34. 10.1103/physrevb.37.6991 / Phys. Rev. B (1988)
  35. 10.1103/physrevb.34.1304 / Phys. Rev. B (1986)
  36. 10.1021/jp004368u / J. Phys. Chem. A (2001)
  37. 10.1021/ct800066q / J. Chem. Theory Comput. (2008)
  38. 10.1021/ct400953f / J. Chem. Theory Comput. (2014)
  39. 10.1039/c5cp02580j / Phys. Chem. Chem. Phys. (2015)
  40. 10.1039/c5cp07319g / Phys. Chem. Chem. Phys. (2016)
  41. 10.1021/acs.jpca.7b02950 / J. Phys. Chem. A (2017)
  42. 10.1002/jcc.20864 / J. Comput. Chem. (2008)
  43. 10.1147/rd.441.0206 / IBM J. Res. Dev. (2000)
  44. 10.1103/physrevlett.108.058301 / Phys. Rev. Lett. (2012)
  45. 10.1088/1367-2630/15/9/095003 / New J. Phys. (2013)
  46. 10.1021/ct400195d / J. Chem. Theory Comput. (2013)
  47. 10.1021/acs.jpclett.5b00831 / J. Phys. Chem. Lett. (2015)
  48. 10.1007/bf02478259 / Bull. Math. Biophys. (1943)
  49. 10.1016/0893-6080(88)90020-2 / Neural Networks (1988)
  50. 10.1142/s0218339094000179 / J. Biol. Syst. (1994)
  51. {'volume-title': 'Neural Networks for Pattern Recognition', 'year': '1995', 'key': '2023080221481817600_c51'} / Neural Networks for Pattern Recognition (1995)
  52. {'first-page': '1', 'volume-title': 'Scientific Applications of Neural Nets', 'year': '1999', 'key': '2023080221481817600_c52'} / Scientific Applications of Neural Nets (1999)
  53. {'volume-title': 'Pattern Recognition and Neural Networks', 'year': '2007', 'key': '2023080221481817600_c53'} / Pattern Recognition and Neural Networks (2007)
  54. {'volume-title': 'Neural Networks and Learning Machines', 'year': '2009', 'key': '2023080221481817600_c54'} / Neural Networks and Learning Machines (2009)
  55. 10.1109/msp.2012.2205597 / IEEE Signal Process. Mag. (2012)
  56. K. Simonyan and A. Zisserman, preprint arXiv:1409.1556 (2014).
  57. 10.1109/72.554195 / IEEE Trans. Neural Networks (1997)
  58. 10.1007/bf02551274 / Math. Control, Signals, Syst. (1989)
  59. 10.1016/0893-6080(91)90009-t / Neural Networks (1991)
  60. 10.1039/c1cp21668f / Phys. Chem. Chem. Phys. (2011)
  61. 10.1063/1.2336223 / J. Chem. Phys. (2006)
  62. 10.1063/1.2746846 / J. Chem. Phys. (2007)
  63. 10.1063/1.3124802 / J. Chem. Phys. (2009)
  64. 10.1021/jp9105585 / J. Phys. Chem. A (2010)
  65. 10.1021/acs.jpclett.7b01072 / J. Phys. Chem. Lett. (2017)
  66. 10.1103/physrevlett.98.146401 / Phys. Rev. Lett. (2007)
  67. 10.1016/j.cpc.2016.05.010 / Comput. Phys. Commun. (2016)
  68. 10.1103/physrevb.96.014112 / Phys. Rev. B (2017)
  69. 10.1103/physrevb.87.184115 / Phys. Rev. B (2013)
  70. 10.1002/qua.24954 / Int. J. Quantum Chem. (2015)
  71. 10.1002/anie.201703114 / Angew. Chem., Int. Ed. (2017)
  72. 10.1088/0953-8984/26/18/183001 / J. Phys.: Condens. Matter (2014)
  73. 10.1038/ncomms13890 / Nat. Commun. (2017)
  74. {'first-page': '1631', 'article-title': 'Recursive deep models for semantic compositionality over a sentiment treebank', 'year': '2013', 'key': '2023080221481817600_c74'} / Recursive deep models for semantic compositionality over a sentiment treebank (2013)
  75. {'first-page': '1017', 'article-title': 'Generating text with recurrent neural networks', 'year': '2011', 'key': '2023080221481817600_c75'} / Generating text with recurrent neural networks (2011)
  76. {'first-page': '926', 'article-title': 'Reasoning with neural tensor networks for knowledge base completion', 'year': '2013', 'key': '2023080221481817600_c76'} / Reasoning with neural tensor networks for knowledge base completion (2013)
  77. 10.1038/sdata.2014.22 / Sci. Data (2014)
  78. {'key': '2023080221481817600_c78', 'first-page': '1', 'article-title': 'SchNet: A continuous-filter convolutional neural network for modeling quantum interactions', 'volume': '30', 'year': '2017', 'journal-title': 'Adv. Neural Inf. Process. Syst.'} / Adv. Neural Inf. Process. Syst. / SchNet: A continuous-filter convolutional neural network for modeling quantum interactions (2017)
  79. 10.1109/cvpr.2016.90 / IEEE Pattern Recognit. / Deep residual learning for image recognition (2016)
  80. 10.1126/sciadv.1603015 / Sci. Adv. (2017)
  81. 10.1021/ct0501093 / J. Chem. Theory Comput. (2005)
  82. 10.1021/ct0502209 / J. Chem. Theory Comput. (2006)
  83. 10.1021/ct800503d / J. Chem. Theory Comput. (2009)
  84. {'volume-title': 'Digital Design and Computer Architecture', 'year': '2010', 'key': '2023080221481817600_c85'} / Digital Design and Computer Architecture (2010)
  85. {'key': '2023080221481817600_c86', 'first-page': '3111', 'article-title': 'Distributed representations of words and phrases and their compositionality', 'volume': '26', 'year': '2013', 'journal-title': 'Adv. Neural Inf. Process. Syst.'} / Adv. Neural Inf. Process. Syst. / Distributed representations of words and phrases and their compositionality (2013)
  86. {'key': '2023080221481817600_c87', 'first-page': '2265', 'volume': '8', 'year': '2007', 'journal-title': 'J. Mach. Learn. Res.'} / J. Mach. Learn. Res. (2007)
  87. 10.1137/080743627 / SIAM Rev. (2010)
  88. 10.1038/35016072 / Nature (2000)
  89. {'key': '2023080221481817600_c91', 'first-page': '472', 'article-title': 'Incorporating second-order functional knowledge for better option pricing', 'volume': '14', 'year': '2001', 'journal-title': 'Adv. Neural Inf. Process. Syst.'} / Adv. Neural Inf. Process. Syst. / Incorporating second-order functional knowledge for better option pricing (2001)
  90. {'volume-title': 'Neural Networks: Tricks of the Trade', 'year': '2003', 'key': '2023080221481817600_c92'} / Neural Networks: Tricks of the Trade (2003)
  91. {'volume-title': 'Radial Basis Functions, Multi-Variable Functional Interpolation and Adaptive Networks', 'year': '1988', 'key': '2023080221481817600_c93'} / Radial Basis Functions, Multi-Variable Functional Interpolation and Adaptive Networks (1988)
  92. {'key': '2023080221481817600_c94', 'first-page': '321', 'volume': '2', 'year': '1988', 'journal-title': 'Complex Syst.'} / Complex Syst. (1988)
  93. 10.1016/s0893-6080(01)00027-2 / Neural Networks (2001)
  94. G. Klambauer, T. Unterthiner, A. Mayr, and S. Hochreiter, in Advance in Neural Information Processing Systems, edited by I. Guyon et al. (Curran Associates, Inc., 2017), Vol. 30, pp. 971–980.
  95. {'first-page': '249', 'article-title': 'Understanding the difficulty of training deep feedforward neural networks', 'year': '2010', 'key': '2023080221481817600_c97'} / Understanding the difficulty of training deep feedforward neural networks (2010)
  96. 10.1021/ci300415d / J. Chem. Inf. Model. (2012)
  97. 10.1103/physrevlett.77.3865 / Phys. Rev. Lett. (1996)
  98. 10.1103/physrevlett.102.073005 / Phys. Rev. Lett. (2009)
  99. 10.1002/jcc.21287 / J. Comput. Chem. (2009)
  100. 10.1063/1.3447701 / J. Chem. Phys. (2010)
  101. {'year': '2009', 'key': '2023080221481817600_c103'} (2009)
  102. D. Kingma and J. Ba, preprint arXiv:1412.6980 (2014).
  103. {'volume-title': 'Advanced Engineering Mathematics', 'year': '2011', 'key': '2023080221481817600_c105'} / Advanced Engineering Mathematics (2011)
  104. {'year': '2015', 'key': '2023080221481817600_c106', 'article-title': 'TensorFlow: Large-scale machine learning on heterogeneous systems'} / TensorFlow: Large-scale machine learning on heterogeneous systems (2015)
  105. 10.1111/j.2517-6161.1991.tb01857.x / J. R. Stat. Soc., Ser. B (1991)
  106. {'volume-title': 'Compendium of Chemical Terminology', 'year': '1997', 'key': '2023080221481817600_c108'} / Compendium of Chemical Terminology (1997)
  107. {'volume-title': 'SMILES: A Line Notation and Computerized Interpreter for Chemical Structures', 'year': '1987', 'key': '2023080221481817600_c109'} / SMILES: A Line Notation and Computerized Interpreter for Chemical Structures (1987)
  108. {'volume-title': 'Graph Algorithms', 'year': '2011', 'key': '2023080221481817600_c110'} / Graph Algorithms (2011)
  109. 10.1103/physrev.77.717 / Phys. Rev. (1950)
  110. 10.1021/ed049p400 / J. Chem. Educ. (1972)
  111. {'volume-title': 'Interpretation of Carbon-13 NMR Spectra', 'year': '1988', 'key': '2023080221481817600_c113'} / Interpretation of Carbon-13 NMR Spectra (1988)
  112. {'volume-title': 'Lange’s Handbook of Chemistry', 'year': '1967', 'key': '2023080221481817600_c114'} / Lange’s Handbook of Chemistry (1967)
  113. 10.1038/s41467-017-00839-3 / Nat. Commun. (2017)
  114. 10.1103/physrevb.83.153101 / Phys. Rev. B (2011)
  115. 10.1007/bf00549096 / Theor. Chim. Acta (1977)
  116. 10.1021/jp035748t / J. Phys. Chem. A (2003)
  117. 10.1002/jcc.22996 / J. Comput. Chem. (2012)
  118. M. Gastegger, J. Behler, and P. Marquetand, Chem. Sci. 8, 6924–6935 (2017).10.1039/C7SC02267K (10.1039/C7SC02267K)
  119. 10.1021/jp401225b / J. Phys. Chem. A (2013)
  120. 10.1063/1.3382344 / J. Chem. Phys. (2010)
  121. 10.1073/pnas.1602375113 / Proc. Natl. Acad. Sci. U. S. A. (2016)
Dates
Type When
Created 7 years, 5 months ago (March 15, 2018, 11:34 a.m.)
Deposited 1 year, 1 month ago (July 1, 2024, 9:36 p.m.)
Indexed 3 weeks, 5 days ago (July 30, 2025, 7:11 a.m.)
Issued 7 years, 5 months ago (March 15, 2018)
Published 7 years, 5 months ago (March 15, 2018)
Published Online 7 years, 5 months ago (March 15, 2018)
Published Print 7 years, 1 month ago (June 28, 2018)
Funders 1
  1. Swiss National Science Foundation 10.13039/501100001711 Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung

    Region: Europe

    pri (Trusts, charities, foundations (both public and private))

    Labels10
    1. Schweizerischer Nationalfonds
    2. Swiss National Science Foundation
    3. Fonds National Suisse de la Recherche Scientifique
    4. Fondo Nazionale Svizzero per la Ricerca Scientifica
    5. Fonds National Suisse
    6. Fondo Nazionale Svizzero
    7. Schweizerische Nationalfonds
    8. SNF
    9. SNSF
    10. FNS
    Awards1
    1. 200021-7117810

@article{Unke_2018, title={A reactive, scalable, and transferable model for molecular energies from a neural network approach based on local information}, volume={148}, ISSN={1089-7690}, url={http://dx.doi.org/10.1063/1.5017898}, DOI={10.1063/1.5017898}, number={24}, journal={The Journal of Chemical Physics}, publisher={AIP Publishing}, author={Unke, Oliver T. and Meuwly, Markus}, year={2018}, month=mar }