Toward High-level Machine Learning Potential for Water Based on Quantum Fragmentation and Neural Networks

To Pair or not to Pair? Machine-Learned Explicitly-Correlated Electronic Structure for NaCl in Water

The extent of ion pairing in solution is an important phenomenon to rationalize transport and thermodynamic properties of electrolytes. A fundamental measure of this pairing is the potential of mean force (PMF) between solvated ions. The relative stabilities of the paired and solvent shared states in the PMF and the barrier between them are highly sensitive to the underlying potential energy surface. However, direct application of accurate electronic structure methods is challenging, since long simulations are required. We develop wave function based machine learning potentials with the random phase approximation (RPA) and second order Møller-Plesset (MP2) perturbation theory for the prototypical system of Na and Cl ions in water. We show both methods in agreement, predicting the paired and solvent shared states to have similar energies (within 0.2 kcal/mol). We also provide the same benchmarks for different DFT functionals as well as insight into the PMF based on simple analyses of the interactions in the system.

Machine Learning Potentials with the Iterative Boltzmann Inversion: Training to Experiment

Methodologies for training machine learning potentials (MLPs) with quantum-mechanical simulation data have recently seen tremendous progress. Experimental data have a very different character than simulated data, and most MLP training procedures cannot be easily adapted to incorporate both types of data into the training process. We investigate a training procedure based on iterative Boltzmann inversion that produces a pair potential correction to an existing MLP using equilibrium radial distribution function data. By applying these corrections to an MLP for pure aluminum based on density functional theory, we observe that the resulting model largely addresses previous overstructuring in the melt phase. Interestingly, the corrected MLP also exhibits improved performance in predicting experimental diffusion constants, which are not included in the training procedure. The presented method does not require autodifferentiating through a molecular dynamics solver and does not make assumptions about the MLP architecture. Our results suggest a practical framework for incorporating experimental data into machine learning models to improve the accuracy of molecular dynamics simulations.

A Status Report on "Gold Standard" Machine-Learned Potentials for Water

Owing to the central importance of water to life as well as its unusual properties, potentials for water have been the subject of extensive research over the past 50 years. Recently, five potentials based on different machine learning approaches have been reported that are at or near the "gold standard" CCSD(T) level of theory. The development of such high-level potentials enables efficient and accurate simulations of water systems using classical and quantum dynamical approaches. This Perspective serves as a status report of these potentials, focusing on their methodology and applications to water systems across different phases. Their performances on the energies of gas phase water clusters, as well as condensed phase structural and dynamical properties, are discussed.

Accurate SCC-DFTB Parametrization of Liquid Water with Improved Atomic Charges and Iterative Boltzmann Inversion

This work presents improvements of the description of liquid water within the self-consistent-charge density-functional based tight-binding scheme combining the use of Weighted Mulliken (WMull) charges and optimized O-H repulsive potential through the iterative Boltzmann inversion (IBI) process. The quality of the newly developed models is validated considering pair radial distribution functions (RDFs), as well as other structural, energetic, thermodynamic, and dynamic properties. The use of WMull charges certainly improves the agreement with experimental data, however leading to over-structured RDFs at short distance, that can be further improved by considering an optimized O-H repulsive potential obtained by the IBI process. Three different schemes were used to optimize this potential: (i) optimization including short O-H distances. This led to accurate RDFs as well as improved self-diffusion coefficient and heat of vaporization, while the proton transfer energy barrier is severely deteriorated; (ii) optimization starting at long distance. The proton transfer energy barrier is recovered while the heat of vaporization is deteriorated and the O-H RDF is less accurate at short distance; (iii) optimization within the path-integral molecular dynamics scheme which allows us to exclude nuclear quantum effects from the repulsive potential. The latter potential, in conjunction with the WMull improved atomic charges, provides similar results as (i) for structural, dynamic, and thermodynamic properties while recovering a large part of the proton transfer energy barrier. It therefore offers a good compromise to study both dynamic properties and chemistry within liquid water at a quantum chemical level.

Data-Efficient Machine Learning Potentials from Transfer Learning of Periodic Correlated Electronic Structure Methods: Liquid Water at AFQMC, CCSD, and CCSD(T) Accuracy

Obtaining the atomistic structure and dynamics of disordered condensed-phase systems from first-principles remains one of the forefront challenges of chemical theory. Here we exploit recent advances in periodic electronic structure and provide a data-efficient approach to obtain machine-learned condensed-phase potential energy surfaces using AFQMC, CCSD, and CCSD(T) from a very small number (≤200) of energies by leveraging a transfer learning scheme starting from lower-tier electronic structure methods. We demonstrate the effectiveness of this approach for liquid water by performing both classical and path integral molecular dynamics simulations on these machine-learned potential energy surfaces. By doing this, we uncover the interplay of dynamical electron correlation and nuclear quantum effects across the entire liquid range of water while providing a general strategy for efficiently utilizing periodic correlated electronic structure methods to explore disordered condensed-phase systems.

Microscopic Mechanism of Proton Transfer in Pure Water under Ambient Conditions

Water molecules and the associated proton transfer (PT) are prevalent in chemical and biological systems and have been a hot research topic. Spectroscopic characterization and ab initio molecular dynamics (AIMD) simulations have previously revealed insights into acidic and basic liquids. Presumably, the situation in the acidic/basic solution is not necessarily the same as in pure water; in addition, the autoionization constant for water is only 10^-14 under ambient conditions, making the study of PT in pure water challenging. To overcome this issue, we modeled periodic water box systems containing 1000 molecules for tens of nanoseconds based on a neural network potential (NNP) with quantum mechanical accuracy. The NNP was generated by training a dataset containing the energies and atomic forces of 17 075 configurations of periodic water box systems, and these data points were calculated at the MP2 level that considers electron correlation effects. We found that the size of the system and the duration of the simulation have a significant impact on the convergence of the results. With these factors considered, our simulations showed that hydronium (H₃O⁺) and hydroxide (OH^-) ions in water have distinct hydration structures, thermodynamic and kinetic properties, e.g., the longer-lasting and more stable hydrated structure of OH^- ions than that of H₃O⁺, as well as a significantly higher free energy barrier for the OH^--associated PT than that of H₃O⁺, leading the two to exhibit completely different PT behaviors. Given these characteristics, we further found that PT via OH^- ions tends not to occur multiple times or between many molecules. In contrast, PT via H₃O⁺ can synergistically occur among multiple molecules and prefers to adopt a cyclic pattern among three water molecules, while it occurs mostly in a chain pattern when more water molecules are involved. Therefore, our studies provide a detailed and solid microscopic explanation for the PT process in pure water.

Hydrogen Bonding and Infrared Spectra of Ethyl-3-methylimidazolium Bis(trifluoromethylsulfonyl)imide/Water Mixtures: A View from Molecular Dynamics Simulations

Simulations of ab initio molecular dynamics have been performed for mixtures of ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMIM-TFSI) ionic liquid and water. Statistics of donors and acceptors of hydrogen bonds has revealed that with increasing water content, hydrogen bonds between EMIM cations and TFSI anions are replaced by bonds to water molecules. In the mixture of liquids, the total number of bonds (from EMIM cations or water molecules) formed by TFSI acceptors increases. IR spectra obtained from ab initio molecular dynamics trajectories are in good agreement with literature data for ionic liquid/water systems. Analysis of oscillations of individual C-H and O-H bonds has shown correlations between vibrational frequencies and hydrogen bonds formed by an EMIM cation or water molecule and has indicated that the changes in the IR spectrum result from the decreased number of water-water hydrogen bonds in the mixture. The tests of DFTB methodology with tailored parameterizations have yielded reasonably good description of the IR spectrum of bulk water, whereas available parameterizations have failed in satisfactory reproduction of the IR spectrum of EMIM-TFSI/water mixtures in the region above 3000 cm^-1.