The Latent Ewald Summation (LES) framework infers atomic charges and the resulting long-range electrostatics directly from energy and force data. The charges can reproduce polarization and Born effective charges (BECs), and be used to drive MD simulations under electric fields.
[1] The original method paper:
https://www.nature.com/articles/s41524-025-01577-7
[2] Besides learning long-range interactions, LES can learn physical charges on each atom:
https://arxiv.org/abs/2412.15455
[3] LES can infer how the system responds to electric field. This is important for modeling electrochemical systems, and can be used to predict infrared spectra.
https://arxiv.org/abs/2504.05169
[4] The LES algorithm works for any machine learning interatomic potential. One can make a foundation model using LES.
Chemical potentials are fundamental but difficult to compute. We developed the S0 method that is able to extract chemical potentials from equilibrium molecular dynamics (MD) simulations. This means one can run a standard MD for mixtures, and determine chemical potentials using a simple post-processing step.
[1] The S0 paper: https://pubs.aip.org/aip/jcp/article/157/12/121101/2841712
[2] Solubility of molecular crystals: https://pubs.aip.org/aip/jcp/article/159/18/184110/2921030
[3] Adsorption of gas in porous materials: https://pubs.aip.org/aip/jcp/article/158/16/161101/2884975
[4] Azeotropes: https://pubs.aip.org/aip/jcp/article/161/3/034111/3303346
[5] Liquid-liquid phase separation of hydrocarbon mixtures in planets: https://www.nature.com/articles/s41467-023-36841-1
Our method to perform ab initio thermodynamics is to use machine learning methods to learn the atomic interactions from quantum mechanics, and thus avoiding otherwise prohibitively expensive calculations that explicitly consider electrons.
In [1], we computed the first phase diagram of water at ab intio level of theory, accounting for thermal and nuclear fluctuations as well as proton disorder. The computed phase diagrams are in qualitative agreement with experiment.
In [2], we computed the phase diagram of dense hydrogen. We found evidence for a continuous molecular to atomic transition in the dense hydrogen fluid, instead of a first-order one. The transition is smooth because the associated critical point is hidden under the melting line of hydrogen. Furthermore, this hidden critical point also induces other unusual phenomena, including density and heat capacity maxima. In [3], we computed the phase diagram of superionic water inside ice giants such as Uranus and Neptune. Our phase boundaries help resolve the fractions of insulating ice, different superionic phases and liquid water inside the ice giants.
[1] Aleks Reinhardt*, Bingqing Cheng*. (2021) Quantum-mechanical exploration of the phase diagram of water. Nature communications 12.1: 1-7.
[2] Bingqing Cheng*, Guglielmo Mazzola, Chris J. Pickard, Michele Ceriotti. (2020) Evidence for supercritical behaviour of high-pressure liquid hydrogen. Nature, 585, 217–220
(see the Cambridge Research and The Economist features)
[3] Bingqing Cheng*, Mandy Bethkenhagen, Chris J. Pickard, Sebastien Hamel. (2021) Phase behaviours of superionic water at planetary conditions. Nature Physics,17, 1228-1232.
The pipe dream of any theoretician is to model any material and predict its properties from the laws of quantum mechanics. The difficulty here comes from the high computational cost of applying these laws, particularly when modelling systems larger than hundreds of atoms at a finite temperature.
Our method to perform ab initio thermodynamics is to use machine learning methods to learn the atomic interactions from quantum mechanics, and thus avoiding otherwise prohibitively expensive calculations that explicitly consider electrons. In this study of water[1], not only are we able to reproduce closely the experimental values of many of the properties of water, but we can also rationalize phenomena including the density anomaly of water, the floating of ice, the difference between the melting point of heavy and normal water, and the six-fold symmetry of snowflakes, by identifying their physical origin.
[1] Bingqing Cheng*, Edgar A Engel, Jörg Behler, Christoph Dellago, Michele Ceriotti. (2019) ab initio thermodynamics of liquid and solid water. Proceedings of the National Academy of Sciences, 116 (4), 1110-1115.
Nowadays, a typical dataset in computational physics, chemistry and materials science contains thousands to millions of atomic structures, along with a diverse range of properties. It is thus be desirable to have a data-driven and automated framework for visualizing and analyzing such structural datasets.
In [1], we describe how to construct a low-dimensional representation of the datasets of materials and molecules. To largely automate the process, we have developed user-friendly software packages: ASAP [2] is a Python-based command-line tool that enables automatic analysis and mapping. [3] shows an example on how we apply the framework to crystal structure predictions. In [4], we revealed that the local environments characterizing the different ice phases are all present in the liquid water, which implies that water models created to describe the liquid can be transferred to study the ices.
[1] Bingqing Cheng*, et al. (2020) Mapping Materials and Molecules. Accounts of Chemical Research, 12697-12705.
[2] https://github.com/BingqingCheng/ASAP
[3] Aleks Reinhardt, Chris J. Pickard, Bingqing Cheng*. (2020) Predicting the phase diagram of titanium dioxide with random search and pattern recognition. Physical Chemistry Chemical Physics, 22: 12697–12705.
[4] Bartomeu Monserrat, Jan Gerit Brandenburg, Edgar A Engel, Bingqing Cheng*. (2020) Liquid water contains the building blocks of diverse ice phases. Nature Communications 11.1: 1-8.
(see Chemistry Community blog post)
The thermal conductivity of a fluid measures how well it conducts heat. Understanding the heat transport process is not only fundamentally important, but also has technological implications in material manufacturing, thermoelectric conversion, energy saving, heat dissipation, and many more.
In [1], we proposed a method to compute the thermal conductivity of a bulk fluid unambiguously, just using equilibrium molecular dynamics trajectories. This method thus bypasses the conceptual and practial difficulties of using the conventional Green-Kubo method. No heat flux, no energy, no force, no Green-Kubo, no temperature gradient, no problem!
[1] Bingqing Cheng*, Daan Frenkel (2020) Computing the Heat Conductivity of Fluids from Density Fluctuations. Physical Review Letters, 125, 130602 (Editor's suggestion)
It is well-known that electrons have to be described quantum-mechanically. On the other hand, since atomic nuclei are much heavier, they are often treated as classical particles in atomistic simulations. However, for light elements such as hydrogen and lithium the classical approximation can break down , and nuclear quantum effects (NQEs) can play a crucial role in numerous phenomena such as the isotope effect, hydrogen transfer rates, and the heat capacity of solids.
We consider NQEs using a path-integral formalism, which exploits an isomorphism between a quantum-mechanical nucleus and a ring-polymer. We have studied NQEs related problems including isotope fractionations [1], kinetic energies of water [2], and hydrogen embrittlement[3].
[1] Bingqing Cheng, Michele Ceriotti. (2014) Direct path integral estimators for isotope fractionation ratios. The Journal of Chemical Physics, 141(24): 244112.
[2] Bingqing Cheng, Jörg Behler, Michele Ceriotti. (2016) Nuclear Quantum Effects in Water at the Triple Point: Using Theory as a Link Between Experiments. Journal of Physical Chemistry Letters, 7(12): 2210-2215.
[3] Bingqing Cheng, Anthony T Paxton, Michele Ceriotti. (2018) Hydrogen diffusion and trapping in a-iron: the role of quantum and anharmonic fluctuations. Physical Review Letters, 120(22): 225901.
When a nucleus grows inside a bulk phase, free energy is gained by the interior, while the interface incurs a penalty. The competition between the two contributions results in a free energy barrier, which the system must overcome for the nucleus to grow to a critical size ultimately leading to an avalanche of structural transitions.
Atomistic modeling is a powerful tool for capturing the dynamical processes and investigating the underlying mechanism of nucleation, but it is difficult to link between the atomistic picture with macroscopic nucleation theories.
Inspired by the Gibbs dividing surface proposed by one of the founding fathers of statistical mechanics, we formulated a thermodynamic framework that reconciles the atomisic picture with macroscopic theories of nucleation [1-2]. Crucially, by defining the interface between the two phases in a rigorous and self-consistent manner, we are able to extract accurately the free energy associated with the nucleus and the interface, enabling stringent validation and extension of macroscopic nucleation theories [3-5].
[1] Bingqing Cheng, Gareth A Tribello, Michele Ceriotti. (2015) Solid-liquid interfacial free energy out of equilibrium. Physical Review B, 92(18): 180102.
[2] Bingqing Cheng*, Michele Ceriotti. (2017) Bridging the gap between atomistic and macroscopic models of homogeneous nucleation. The Journal of Chemical Physics, 146(3): 034106.
[3] Bingqing Cheng*, Gareth A Tribello, Michele Ceriotti. (2017) The Gibbs free energy of homogeneous nucleation: from atomistic nuclei to the planar limit. The Journal of Chemical Physics, 147(10): 104707.
[4] Bingqing Cheng*, Michele Ceriotti. (2018) Computing the Tolman length for solid-liquid interfaces. The Journal of Chemical Physics, 148(23): 231102.
[5] Bingqing Cheng*, Michele Ceriotti, Gareth A Tribello (2020) Classical nucleation theory predicts the shape of the nucleus in homogeneous solidification. The Journal of chemical physics, 152(4), 044103.
