+
- Physical Review Journals
All JournalsPhysics Magazine

Physical Review Research

Reuse & Permissions

It is not necessary to obtain permission to reuse this article or its components as it is available under the terms of the Creative Commons Attribution 4.0 International license. This license permits unrestricted use, distribution, and reproduction in any medium, provided attribution to the author(s) and the published article's title, journal citation, and DOI are maintained. Please note that some figures may have been included with permission from other third parties. It is your responsibility to obtain the proper permission from the rights holder directly for these figures.

  • Open Access

Ab initio solution of the many-electron Schrödinger equation with deep neural networks

David Pfau*,†, James S. Spencer*, and Alexander G. D. G. Matthews

W. M. C. Foulkes

  • DeepMind, 6 Pancras Square, London N1C 4AG, United Kingdom

  • Department of Physics, Imperial College London, South Kensington Campus, London SW7 2AZ, United Kingdom
  • *These authors contributed equally to this work.
  • †pfau@google.com
PDF

Phys. Rev. Research 2, 033429 – Published 16 September, 2020

DOI: https://doi.org/10.1103/PhysRevResearch.2.033429

Abstract

Given access to accurate solutions of the many-electron Schrödinger equation, nearly all chemistry could be derived from first principles. Exact wave functions of interesting chemical systems are out of reach because they are NP-hard to compute in general, but approximations can be found using polynomially scaling algorithms. The key challenge for many of these algorithms is the choice of wave function approximation, or Ansatz, which must trade off between efficiency and accuracy. Neural networks have shown impressive power as accurate practical function approximators and promise as a compact wave-function Ansatz for spin systems, but problems in electronic structure require wave functions that obey Fermi-Dirac statistics. Here we introduce a novel deep learning architecture, the Fermionic neural network, as a powerful wave-function Ansatz for many-electron systems. The Fermionic neural network is able to achieve accuracy beyond other variational quantum Monte Carlo Ansatz on a variety of atoms and small molecules. Using no data other than atomic positions and charges, we predict the dissociation curves of the nitrogen molecule and hydrogen chain, two challenging strongly correlated systems, to significantly higher accuracy than the coupled cluster method, widely considered the most accurate scalable method for quantum chemistry at equilibrium geometry. This demonstrates that deep neural networks can improve the accuracy of variational quantum Monte Carlo to the point where it outperforms other ab initio quantum chemistry methods, opening the possibility of accurate direct optimization of wave functions for previously intractable many-electron systems.

Physics Subject Headings (PhySH)

Article Text

References (73)

  1. A. Krizhevsky, I. Sutskever, and G. E. Hinton, in Neural Information Processing Systems (MIT Press, Cambridge, MA, 2012), Vol. 25, pp. 1097–1105.
  2. D. Silver, A. Huang, C. J. Maddison, A. Guez, L. Sifre, G. Van Den Driessche, J. Schrittwieser, I. Antonoglou, V. Panneershelvam, M. Lanctot et al., Nature 529, 484 (2016).
  3. J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl, in Proceedings of the 34th International Conference on Machine Learning (ICML) (JMLR.org, 2017), Vol. 70, pp. 1263–1272.
  4. K. T. Schütt, M. Gastegger, A. Tkatchenko, K.-R. Müller, and R. J. Maurer, Nat. Commun. 10, 5024 (2019).
  5. K. Mills, M. Spanner, and I. Tamblyn, Phys. Rev. A 96, 042113 (2017).
  6. A. V. Sinitskiy and V. S. Pande, arXiv:1908.00971 (2019).
  7. L. Cheng, M. Welborn, A. S. Christensen, and T. F. Miller III, J. Chem. Phys. 150, 131103 (2019).
  8. R. J. Bartlett and M. Musiał, Rev. Modern Phys. 79, 291 (2007).
  9. M. Troyer and U. J. Wiese, Phys. Rev. Lett. 94, 170201 (2005).
  10. M. Born and R. Oppenheimer, Ann. Phys. 389, 457 (1927).
  11. G. H. Booth and A. Alavi, J. Chem. Phys. 132, 174104 (2010).
  12. W. M. C. Foulkes, L. Mitas, R. J. Needs, and G. Rajagopal, Rev. Modern Phys. 73, 33 (2001).
  13. R. P. Feynman and M. Cohen, Phys. Rev. 102, 1189 (1956).
  14. M. Bajdich, L. Mitas, G. Drobný, L. K. Wagner, and K. E. Schmidt, Phys. Rev. Lett. 96, 130201 (2006).
  15. R. Orús, Ann. Phys. 349, 117 (2014).
  16. G. Carleo and M. Troyer, Science 355, 602 (2017).
  17. K. Choo, G. Carleo, N. Regnault, and T. Neupert, Phys. Rev. Lett. 121, 167204 (2018).
  18. A. Nagy and V. Savona, Phys. Rev. Lett. 122, 250501 (2019).
  19. Y. Nomura, A. S. Darmawan, Y. Yamaji, and M. Imada, Phys. Rev. B 96, 205152 (2017).
  20. L. Yang, Z. Leng, G. Yu, A. Patel, W.-J. Hu, and H. Pu, Phys. Rev. Research 2, 012039 (2020).
  21. D. Luo and B. K. Clark, Phys. Rev. Lett. 122, 226401 (2019).
  22. H. Saito, J. Phys. Soc. Jpn. 87, 074002 (2018).
  23. J. Kessler, C. Calcavecchia, and T. D. Kühne, arXiv:1904.10251 (2019).
  24. J. Han, L. Zhang, and W. E, J. Comput. Phys. 399, 108929 (2019).
  25. M. Taddei, M. Ruggeri, S. Moroni, and M. Holzmann, Phys. Rev. B 91, 115106 (2015).
  26. M. Ruggeri, S. Moroni, and M. Holzmann, Phys. Rev. Lett. 120, 205302 (2018).
  27. Since this manuscript appeared online, several other works using neural networks as Ansatz for continuous-space fermionic systems have appeared [72, 73]. The first [72] also augments the typical Slater-Jastrow Ansatz with a deep neural network, while physical constraints like the cusp conditions are included explicitly. As the model has fewer parameters than ours, it is faster to optimize, but does not achieve the same accuracy. The second [73] represents a chemical system in second-quantized form using a given basis set, then fits a restricted Boltzmann machine to the ground state. This model is also able to exceed the performance of CCSD(T) within a basis set. However, it is less clear how easily this model extrapolates to the complete basis set limit. Our approach sidesteps the difficulty of choosing a basis set entirely.
  28. S. Zhang, Ab initio electronic structure calculations by auxiliary-field quantum Monte Carlo, in Handbook of Materials Modeling: Methods: Theory and Modeling, edited by W. Andreoni and S. Yip (Springer International Publishing, Berlin, 2018), pp. 1–27.
  29. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, in Advances in Neural Information Processing Systems (NeurIPS) (2017), pp. 5998–6008.
  30. A. W. Senior, R. Evans, J. Jumper, J. Kirkpatrick, L. Sifre, T. Green, C. Qin, A. Žídek, A. W. R. Nelson, A. Bridgland, H. Penedones, S. Petersen, K. Simonyan, S. Crossan, P. Kohli, D. Jones, D. Silver, K. Kavukcuoglu, and D. Hassabis, Nature 577, 706 (2020).
  31. K. T. Schütt, F. Arbabzadah, S. Chmiela, K. R. Müller, and A. Tkatchenko, Nat. Commun. 8, 13890 (2017).
  32. K. T. Schütt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, and K.-R. Müller, J. Chem. Phys. 148, 241722 (2018).
  33. J. Shawe-Taylor, Building symmetries into feedforward networks, First IEE International Conference on Artificial Neural Networks, Conf. Publ. No. 313, London, UK (IEEE, 1989), pp. 158–162.
  34. S. J. Chakravorty, S. R. Gwaltney, E. R. Davidson, F. A. Parpia, and C. F. Fischer, Phys. Rev. A 47, 3649 (1993).
  35. B. K. Clark, M. A. Morales, J. McMinis, J. Kim, and G. E. Scuseria, J. Chem. Phys. 135, 244105 (2011).
  36. E. Neuscamman, C. J. Umrigar, and Garnet Kin-Lic Chan, Phys. Rev. B 85, 045103 (2012).
  37. R. Assaraf, S. Moroni, and C. Filippi, J. Chem. Theory Comput. 13, 5273 (2017).
  38. L. Otis and E. Neuscamman, Phys. Chem. Chem. Phys. 21, 14491 (2019).
  39. I. Sabzevari, A. Mahajan, and S. Sharma, J. Chem. Phys. 152, 024111 (2020).
  40. J. Martens and R. Grosse, in Proceedings of the 32nd International Conference on Machine Learning (ICML) (JMLR.org, 2015), Vol. 37, pp. 2408–2417.
  41. S. Amari, Neural Comput. 10, 251 (1998).
  42. S. Sorella, Phys. Rev. Lett. 80, 4558 (1998).
  43. J. Toulouse and C. Umrigar, J. Chem. Phys. 126, 084102 (2007).
  44. P. Seth, P. L. Ríos, and R. J. Needs, J. Chem. Phys. 134, 084105 (2011).
  45. W. Klopper, R. A. Bachorz, D. P. Tew, and C. Hättig, Phys. Rev. A 81, 022503 (2010).
  46. W. Cencek and J. Rychlewski, Chem. Phys. Lett. 320, 549 (2000).
  47. C. Filippi and C. Umrigar, J. Chem. Phys. 105, 213 (1996).
  48. E. Giner, R. Assaraf, and J. Toulouse, Mol. Phys. 114, 910 (2016).
  49. M. Dash, S. Moroni, A. Scemama, and C. Filippi, J. Chem. Theory Comput. 14, 4176 (2018).
  50. T. Van Voorhis and M. Head-Gordon, J. Chem. Phys. 113, 8873 (2000).
  51. H. Burton and A. Thom, J. Chem. Theory Comput. 12, 167 (2016).
  52. D. Lyakh, M. Musiał, V. Lotrich, and R. Bartlett, Chem. Rev. 112, 182 (2011).
  53. R. J. Le Roy, Y. Huang, and C. Jary, J. Chem. Phys. 125, 164310 (2006).
  54. R. Gdanitz, Chem. Phys. Lett. 283, 253 (1998).
  55. M. Motta, D. M. Ceperley, Garnet Kin-Lic Chan, J. A. Gomez, E. Gull, S. Guo, C. A. Jiménez-Hoyos, T. N. Lan, J. Li, F. Ma, A. J. Millis, N. V. Prokof'ev, U. Ray, G. E. Scuseria, S. Sorella, E. M. Stoudenmire, Q. Sun, I. S. Tupitsyn, S. R. White, D. Zgid, and S. Zhang (Simons Collaboration on the Many-Electron Problem), Phys. Rev. X 7, 031059 (2017).
  56. A. Badinski, P. Haynes, J. Trail, and R. Needs, J. Phys. Condens. Matter 22, 074202 (2010).
  57. M. Motta, C. Genovese, F. Ma, Z.-H. Cui, R. Sawaya, G. K. Chan, N. Chepiga, P. Helms, C. Jimenez-Hoyos, A. J. Millis et al., arXiv:1911.01618 (2019).
  58. R. Q. Hood, M. Y. Chou, A. J. Williamson, G. Rajagopal, R. J. Needs, and W. M. C. Foulkes, Phys. Rev. Lett. 78, 3350 (1997).
  59. Q. Sun, T. C. Berkelbach, N. S. Blunt, G. H. Booth, S. Guo, Z. Li, J. Liu, J. D. McClain, E. R. Sayfutyarova, S. Sharma, S. Wouters, and G. K.-L. Chan, Wiley Interdiscip. Rev.: Comput. Mol. Sci. 8, e1340 (2018).
  60. H. Flyvbjerg and H. G. Petersen, J. Chem. Phys. 91, 461 (1989).
  61. C. Umrigar, M. Nightingale, and K. Runge, J. Chem. Phys. 99, 2865 (1993).
  62. R. M. Lee, G. J. Conduit, N. Nemec, P. López Ríos, and N. D. Drummond, Phys. Rev. E 83, 066706 (2011).
  63. R. Needs, M. Towler, N. Drummond, and P. L. Rios, CASINO: User's Guide Version 2.13 (2015).
  64. R. M. Parrish, L. A. Burns, D. G. A. Smith, A. C. Simmonett, A. E. DePrince, E. G. Hohenstein, U. Bozkaya, A. Y. Sokolov, R. Di Remigio, R. M. Richard, J. F. Gonthier, A. M. James, H. R. McAlexander, A. Kumar, M. Saitow, X. Wang, B. P. Pritchard, P. Verma, H. F. Schaefer, K. Patkowski, R. A. King, E. F. Valeev, F. A. Evangelista, J. M. Turney, T. D. Crawford, and C. D. Sherrill, J. Chem. Theory Comput. 13, 3185 (2017).
  65. A. DePrince and C. Sherrill, J. Chem. Theory Comput. 9, 2687 (2013).
  66. D. Feller, J. Chem. Phys. 96, 6104 (1992).
  67. T. Helgaker and W. Klopper, J. Chem. Phys. 106, 9639 (1997).
  68. D. Petz and C. Sudár, J. Math. Phys. 37, 2662 (1996).
  69. G. Mazzola, A. Zen, and S. Sorella, J. Chem. Phys. 137, 134112 (2012).
  70. M. B. Giles, in Advances in Automatic Differentiation, edited by C. H. Bischof, H. M. Bücker, P. Hovland, U. Naumann, and J. Utke (Springer, Berlin, 2008), pp. 35–44.
  71. L. A. Curtiss, K. Raghavachari, P. C. Redfern, V. Rassolov, and J. Pople, J. Chem. Phys. 109, 7764 (1998).
  72. J. Hermann, Z. Schätzle, and F. Noé, arXiv:1909.08423 (2019).
  73. K. Choo, A. Mezzacapo, and G. Carleo, Nat. Commun. 11, 2368 (2020).

Outline

Information

Phys. Rev. Research 2, 033429– Published 16 September, 2020
  • Received 18 March 2020
  • Accepted 6 August 2020
Crossmark - Check for updates button
Creative Commons Logo

Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article's title, journal citation, and DOI.

Published by the American Physical Society

Sign In to Your Journals Account

Filter

Filter

Article Lookup

Enter a citation

点击 这是indexloc提供的php浏览器服务，不要输入任何密码和下载