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

Dynamical properties of liquid water were studied using Car-Parrinello [Phys. Rev. Lett. 55, 2471 (1985)] ab initio molecular dynamics (AIMD) simulations within the Kohn-Sham (KS) density functional theory employing the Becke-Lee-Yang-Parr exchange-correlation functional for the electronic structure. The KS orbitals were expanded in a discrete variable representation basis set, wherein the complete basis set limit can be easily reached and which, therefore, provides complete convergence of ionic forces. In order to minimize possible nonergodic behavior of the simulated water system in a constant energy (NVE) ensemble, a long equilibration run (30 ps) preceded a 60 ps long production run. The temperature drift during the entire 60 ps trajectory was found to be minimal. The diffusion coefficient [0.055 A2/ps] obtained from the present work for 32 D2O molecules is a factor of 4 smaller than the most up to date experimental value, but significantly larger than those of other recent AIMD studies. Adjusting the experimental result so as to match the finite-sized system used in the present study brings the comparison between theory and experiment to within a factor of 3. More importantly, the system is not observed to become "glassy" as has been reported in previous AIMD studies. The computed infrared spectrum is in good agreement with experimental data, especially in the low frequency regime where the translational and librational motions of water are manifested. The long simulation length also made it possible to perform detailed studies of hydrogen bond dynamics. The relaxation dynamics of hydrogen bonds observed in the present AIMD simulation is slower than those of popular force fields, such as the TIP4P potential, but comparable to that of the TIP5P potential.

Thermodynamic properties of liquid water as well as hexagonal (Ih) and cubic (Ic) ice are predicted based on density functional theory at the hybrid-functional level, rigorously taking into account quantum nuclear motion, anharmonic fluctuations, and proton disorder. This is made possible by combining advanced free-energy methods and state-of-the-art machine-learning techniques. The ab initio description leads to structural properties in excellent agreement with experiments and reliable estimates of the melting points of light and heavy water. We observe that nuclear-quantum effects contribute a crucial [Formula: see text] to the stability of ice Ih, making it more stable than ice Ic. Our computational approach is general and transferable, providing a comprehensive framework for quantitative predictions of ab initio thermodynamic properties using machine-learning potentials as an intermediate step.

The effect of confinement on the dynamical properties of liquid water was studied by mid-infrared ultrafast pump-probe spectroscopy on HDO:D2O in reverse micelles. By preparing water-containing reverse micelles of different well defined sizes, we varied the degree of geometric confinement in water nanodroplets with radii ranging from 0.2 to 4.5 nm. We find that water molecules located near the interface confining the droplet exhibit slower vibrational energy relaxation and have a different spectral absorption than those located in the droplet core. As a result, we can measure the orientational dynamics of these different types of water with high selectivity. We observe that the water molecules in the core show similar orientational dynamics as bulk water and that the water layer solvating the interface is highly immobile.

The properties of high‐pressure liquid water were investigated along the isobar of 25 MPa and in the temperature range 298.15‐623.15 K using classical molecular dynamics simulations. Particular attention has been given to the changes in the local structural and related dynamic properties of liquid water. The results obtained have revealed noticeable changes in the shape of the calculated radial distribution functions, as well as the existence of local extrema or crossovers in several structural descriptors and entropic quantities at temperatures around 423.15 K and 498.15 K, where also significant changes in the hydrogen bond network of liquid water have also been observed. The temperature dependence of translational, reorientational, and hydrogen bond dynamics of liquid water has also been investigated by calculating the translational self‐diffusion and the corresponding O‐H vector Legendre reorientational correlation times and hydrogen bond lifetimes. The corresponding activation energies for each investigated relaxation process have also been presented and discussed.

The Matsuoka-Clementi-Yoshimine (MCY) configuration interaction potential for rigid water-water interactions has been extended to include the intramolecular vibrations. The extended potential (MCYL), using no empirical parameters other than the atomic masses, electron charge, and Planck constant, is used in a molecular-dynamics simulation study of the static and dynamic properties of liquid water. Among the properties studied are internal energy, heat capacity, pressure, radial distribution functions, dielectric constant, static structure factor, velocity autocorrelation functions, self-diffusion coefficients, dipole autocorrelation function, and density and current fluctuations. Comparison with experiments is made whenever possible. Most of these properties are found to improve slightly relative to the MCY model. The simulated high-frequency sound mode seems to support the results and interpretation of a recent coherent inelastic neutron scattering experiment.

Generalized image charge solvation model for electrostatic interactions in molecular dynamics simulations of aqueous solutions

Many remarkable properties of liquid water originate from the ability of its molecules to form hydrogen bonds, each of which emerges as a combination of electrostatic, polarization, dispersion, and donor-acceptor or covalent interactions. In this work, ab initio molecular dynamics was tailored to isolate and switch off the covalent component of interactions between water molecules in simulations. Comparison of simulations with and without covalency shows that a small amount of intermolecular electron density transfer has a profound effect on the structure and dynamics of the hydrogen-bond network and thus on observable properties of room-temperature liquid water.

Coarse-grained water models offer opportunities to study the molecular processes in water on spatial and temporal scales inaccessible to conventional atomic-level simulations. Such a model that can accurately describe the structural, electrostatic and thermodynamic properties of liquid water is difficult to derive but invaluable. Here, we present FlexDQ, a flexible three-site coarse-grained model, representing clusters of water molecules. This model was derived using a full bottom-up approach in which atomistic information of water molecules is processed using clustering techniques based on considerations of both energy and geometry and a mapping algorithm that encodes information of dipole and quadrupole moments of water clusters into coarsely grained interaction sites. Parameterization of FlexDQ was conducted consistently and systematically, augmented by these data. The resulting model reproduces not only dipolar and quadrupolar fluctuations of water clusters, but also their local structural correlations in the bulk phase and, to a good extent, pressure-related properties, dielectric permittivity and water-air interfacial properties that were not targeted for parameterization. The ability of our model to reproduce various properties of water has been shown to originate from its explicit treatment of electrostatics. Coupled with the popular MARTINI CG model, FlexDQ captures the entropy-enthalpy balance of hydrophobic interactions and describes the dielectric behavior of water at the water-membrane interface, yielding a positive internal electrostatic potential in membranes, which is otherwise difficult to reproduce using MARTINI alone. Together, the FlexDQ model and its development framework reported here show promise for studies on the biomolecules in water.

The accuracy of the training data limits the accuracy of bulk properties from machine-learned potentials. For example, hybrid functionals or wave-function-based quantum chemical methods are readily available for cluster data but effectively out of scope for periodic structures. We show that local, atom-centered descriptors for machine-learned potentials enable the prediction of bulk properties from cluster model training data, agreeing reasonably well with predictions from bulk training data. We demonstrate such transferability by studying structural and dynamical properties of bulk liquid water with density functional theory and have found an excellent agreement with experimental and theoretical counterparts.

Water is a unique substance with properties that are not predictable from those of other materials. It has a complex phase space where all of its physical and structural forms possess unexpected properties. Mostly, these are consequences of the ability of the water molecule to form strongly hydrogen‐bonded intricate structures. The extensive hydrogen bonding provides water with the ability to transfer protons and electrons rapidly between the water molecules and to produce positively charged hydrogen ions and negatively charged hydroxide ions. In the many solid phases, each water molecule forms exactly four hydrogen bonds in a tetrahedral arrangement with two hydrogen atoms near each oxygen atom. This arrangement is also present, if much more loosely, in liquid water, where the overall structuring is far more complex. Such liquid water behaves as though it is a mixture of two liquids that change relative composition with variations in temperature and pressure. The properties of liquid water are particularly affected at low temperature when supercooled and at very low temperatures when water glasses are formed. Even gaseous water is affected by this hydrogen bonding.

Evident in many physical and chemical phenomena, thermodynamics is the study of how energy is stored, transformed and transferred in a molecule or material. However, prediction of these properties with simulation techniques is a non-trivial task as several factors such as composition and intermolecular interactions come into play. While molecular dynamics and ab initio molecular dynamics are the most common techniques for the prediction of thermodynamic properties, there exists many shortcomings associated with their use. Therefore, in this work we instead apply QCE theory to predict the thermodynamic properties of liquid water. This theory assumes that a condensed phase system can be represented as a 'mixture' of varying sized clusters rather than as a continuum. As QCE theory relies on first-principle simulations and statistical thermodynamics to determine the thermodynamic behavior of a system, appropriate selection of clusters is a crucial step towards achieving accurate predictions. In this study, we use molecular dynamics and ab initio calculations to obtain initial configurations of 400 water clusters, Wn where n = 3 to 10 and contrast their stability using two different criteria. The role of entropy towards cluster stabilization is investigated by comparing the binding (ΔEBIND/mol) and Gibbs free binding energy per molecule (ΔGBIND/mol) of various Wn at 298.15 K. Initial clustersets are constructed by exploring two-, three-, four and five-combinations of clustersets using the minimum ΔGBIND/mol structures of Wn. We also expand the ΔGBIND/mol criteria for Wn of sizes 3 to 7 to include values larger than 0.0 kJ mol-1 and smaller than 3.0 kJ mol-1 as a means of improving thermodynamic predictions. 37 of the 459 resulting clustersets predicted the correct boiling point of water and its thermodynamic properties with an accuracy of 95%. A scaled population-weighted infrared spectrum was compared to experimental results to validate the composition of the top 5 clustersets. The predicted spectra showed an exact match within the low frequency range (<1000 cm-1) with some discrepancy at the high frequency range (>3400 cm-1). This work highlights that ΔGBIND/mol is so far the best criteria to apply when determining an appropriate clusterset for QCE theory.

Singularity of liquid water-data analysis and modelling

The structural and thermodynamic properties of liquid water and of the dilute solutions of methane and ethane in water were calculated by Monte Carlo simulations in the temperature range 298 K to 318 K and 298 K to 333 K, respectively. The nonpolar molecules were modeled as one- and two-center Lennard–Jones particles; for the interaction potential of water a modified TIP5P model was used. The results indicate that the nonpolar solutes tend to aggregate with increasing temperature. Methane molecules preferably form water-separated pairs, even at higher temperatures, whereas for ethane contact pairs are more likely. For the thermodynamic conditions studied here, the residual chemical potential of water is a linear function of temperature.

Computing dissipative particle dynamics interactions to render molecular structure and temperature-dependent properties of simple liquids

Cryopreservation is the preservation and storage of biomaterialsusing low temperatures. There are several approaches to cryopreservation,and these often include the use of cryoprotectants, which are solutesused to lower the freezing point of water. Isochoric (constant-volume)cryopreservation is a form of cryopreservation that has been gaininginterest over the past 18 years. This method utilizes the anomalousnature of water in that it expands as it freezes. The expansion ofice on freezing is used to induce a pressure in the system that limitsice growth. In this work, we use Gibbsian thermodynamics, the Elliottet al. multisolute osmotic virial equation, the Feistel and Wagnercorrelation for ice Ih, and the Grenke and Elliott correlation forthe thermodynamic properties of liquid water at low temperatures andhigh pressures to predict how the pressure, volume fraction of ice,and solute concentration in the unfrozen fraction change as the solutionis cooled isochorically. We then verified our model by predictingexperimental results for saline solutions and ternary aqueous solutionscontaining NaCl and organic compounds commonly used as cryoprotectants:glycerol, ethylene glycol, propylene glycol, and dimethyl sulfoxide.We found that our model accurately predicts experimental data thatwere collected for cryoprotectant concentrations as high as 5 M, andtemperatures as low as −25 °C. Since we have shown thatour liquid water correlation, on which this work was based, makesaccurate predictions to −70 °C, as long as the pressureis not higher than 400 MPa, we anticipate that the prediction methodspresented in this work will be accurate down to −70 °C.In this work we also modeled how sealing the isochoric chamber atroom temperature versus at the nucleation temperature impacts isochoricfreezing. The prediction methods developed in this work can be usedin the future design of isochoric cryopreservation experiments andprotocols.