Neural Thermodynamic Integration: Free Energies from Energy-Based Diffusion Models
Source abstract
Abstract
Thermodynamic integration (TI) offers a rigorous method for estimating free-energy differences by integrating over a sequence of interpolating conformational ensembles. However, TI calculations are computationally expensive and typically limited to coupling a small number of degrees of freedom due to the need to sample numerous intermediate ensembles with sufficient conformational-space overlap. In this work, we propose to perform TI along an alchemical pathway represented by a trainable neural network, which we term Neural TI. Critically, we parametrize a time-dependent Hamiltonian interpolating between the interacting and noninteracting systems and optimize its gradient using a score matching objective. The ability of the resulting energy-based diffusion model to sample all intermediate ensembles allows us to perform TI from a single reference calculation. We apply our method to Lennard-Jones fluids, where we report accurate calculations of the excess chemical potential, demonstrating that Neural TI reproduces the underlying changes in free energy without the need for simulations at interpolating Hamiltonians.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: Machine Learning in Materials Science · Advanced Thermodynamics and Statistical Mechanics · Phase Equilibria and Thermodynamics
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
Affiliations
Institute for Theoretical Physics, Heidelberg University, 69120 Heidelberg, Germany
Department of Computer Science, University of Geneva, 1227 Carouge, Switzerland
Department of Physics, University of Geneva, 1211 Geneva, Switzerland
Funding
Schweizerischer Nationalfonds zur F?rderung der Wissenschaftlichen Forschung · Deutsche Forschungsgemeinschaft
References (38)
Show source-provided references
1. 10.1021/cr050293k10.1021/cr050293k ↗
2. 10.1146/annurev-biophys-070816-03365410.1146/annurev-biophys-070816-033654 ↗
3. 10.1021/acs.chemrev.1c0052110.1021/acs.chemrev.1c00521 ↗
4. Chipot C. · 2007 · 8610.1007/978-3-540-38448-9 ↗
5. 10.33011/livecoms.2.1.1837810.33011/livecoms.2.1.18378 ↗
6. 10.1126/science.aat266310.1126/science.aat2663 ↗
7. 10.1146/annurev-physchem-042018-05233110.1146/annurev-physchem-042018-052331 ↗
8. 10.1038/s41570-022-00416-310.1038/s41570-022-00416-3 ↗
9. 10.1103/PhysRevE.101.02330410.1103/PhysRevE.101.023304 ↗
10. 10.1103/PhysRevD.100.03451510.1103/PhysRevD.100.034515 ↗
11. 10.1126/science.aaw114710.1126/science.aaw1147 ↗
12. 10.1021/acs.jpclett.2c0332710.1021/acs.jpclett.2c03327 ↗
13. 10.1063/5.001890310.1063/5.0018903 ↗
14. 10.1088/2632-2153/acefa810.1088/2632-2153/acefa8 ↗
15. Dinh L. · 2016
16. Chen R. T. Q. · 2018 · 31
17. Sohl-Dickstein J. · 2015 · 2256
18. Ho J. · Advances in neural information processing systems · 2020 · 33 · 6840
19. Liu Q. · 2016 · 276
20. 10.1063/1.136366810.1063/1.1363668 ↗
21. 10.1063/1.45553910.1063/1.455539 ↗
22. Masrani V. · Advances in Neural Information Processing Systems · 2019 · 32
23. Song Y. · 2020
24. 10.1016/0304-4149(82)90051-510.1016/0304-4149(82)90051-5 ↗
25. 10.1021/acs.jctc.3c0070210.1021/acs.jctc.3c00702 ↗
26. Salimans T. · 2021
27. De Bortoli V. · Advances in Neural Information Processing Systems · 2022 · 35 · 2406
28. Huang C.-W. · Advances in Neural Information Processing Systems · 2022 · 35 · 2750
29. Jing B. · Advances in Neural Information Processing Systems · 2022 · 35 · 24240
30. Máté B. · 2023
31. 10.1016/0009-2614(94)00397-110.1016/0009-2614(94)00397-1 ↗
32. Schütt K. T. · 2017 · 992
33. Gasteiger J. · 2019
34. Batatia I. · Advances in Neural Information Processing Systems · 2022 · 35 · 11423
35. 10.1038/s41467-022-29939-510.1038/s41467-022-29939-5 ↗
36. 10.1063/1.173411010.1063/1.1734110 ↗
37. Frenkel D. · 2023
38. Lipman Y. · 2022