Accurate prediction of ice nucleation from room temperature water

The effect of salt additives on the glycine crystallization pathway revealed by studying one crystal nucleation at a time

Molecular-level understanding of the early stage of crystallization remains a significant challenge. For a wide range of fundamental research and applications, it is of great importance to understand the mechanism of crystal polymorph selections in different solvents or in the presence of additives. We studied glycine crystallization in aqueous solution with or without the addition of sodium chloride (NaCl), one crystal nucleation at a time, using the recently developed single-crystal nucleation spectroscopy (SCNS). It has been reported that glycine forms γ-glycine when salt is added to aqueous solution, and the mechanism has been considered as a classical nucleation where γ-glycine forms directly as a crystal nucleus. Our study shows that metastable polymorph β-glycine forms first in aqueous solution both with or without NaCl through a nonclassical nucleation pathway. NaCl can stabilize the metastable β-glycine over several hours and also prevent it from converting to α-glycine. Eventually γ-glycine nucleates on the surface of β-glycine and then grows while dissolving the mother β-glycine. The crystal habit of β-glycine suggests that the stabilization by NaCl occurs at its polar face. SCNS provides crucial information to accelerate the investigation of the early stage of crystallization toward the rational control of polymorphisms.

Interfacial Ice Density Fluctuations Inform Surface Ice-Philicity

The propensity of a surface to nucleate ice or bind to ice is governed by its ice-philicity─its relative preference for ice over liquid water. However, the relationship between the features of a surface and its ice-philicity is not well understood, and for surfaces with chemical or topographical heterogeneity, such as proteins, their ice-philicity is not even well-defined. In the analogous problem of surface hydrophobicity, it has been shown that hydrophobic surfaces display enhanced low water-density (vapor-like) fluctuations in their vicinity. To interrogate whether enhanced ice-like fluctuations are similarly observed near ice-philic surfaces, here we use molecular simulations and enhanced sampling techniques. Using a family of model surfaces for which the wetting coefficient, k, has previously been characterized, we show that the free energy of observing rare interfacial ice-density fluctuations decreases monotonically with increasing k. By utilizing this connection, we investigate a set of fcc systems and find that the (110) surface is more ice-philic than the (111) or (100) surfaces. By additionally analyzing the structure of interfacial ice, we find that all surfaces prefer to bind to the basal plane of ice, and the topographical complementarity of the (110) surface to the basal plane explains its higher ice-philicity. Using enhanced interfacial ice-like fluctuations as a measure of surface ice-philicity, we then characterize the ice-philicity of chemically heterogeneous and topologically complex systems. In particular, we study the spruce budworm antifreeze protein (sbwAFP), which binds to ice using a known ice-binding site (IBS) and resists engulfment using nonbinding sites of the protein (NBSs). We find that the IBS displays enhanced interfacial ice-density fluctuations and is therefore more ice-philic than the two NBSs studied. We also find the two NBSs are similarly ice-phobic. By establishing a connection between interfacial ice-like fluctuations and surface ice-philicity, our findings thus provide a way to characterize the ice-philicity of heterogeneous surfaces.

Using machine learning with atomistic surface and local water features to predict heterogeneous ice nucleation

Heterogeneous ice nucleation (HIN) has applications in climate science, nanotechnology, and cryopreservation. Ice nucleation on the earth's surface or in the atmosphere usually occurs heterogeneously involving foreign substrates, known as ice nucleating particles (INPs). Experiments identify good INPs but lack sufficient microscopic resolution to answer the basic question: What makes a good INP? We employ molecular dynamics (MD) simulations in combination with machine learning (ML) to address this question. Often, the large amount of computational cost required to cross the nucleation barrier and observe HIN in MD simulations is a practical limitation. We use information obtained from short MD simulations of atomistic surface and water models to predict the likelihood of HIN. We consider 153 atomistic substrates with some surfaces differing in elemental composition and others only in terms of lattice parameters, surface morphology, or surface charges. A range of water features near the surface (local) are extracted from short MD simulations over a time interval (≤300 ns) where ice nucleation has not initiated. Three ML classification models, Random Forest (RF), support vector machine, and Gaussian process classification are considered, and the accuracies achieved by all three approaches lie within their statistical uncertainties. Including local water features is essential for accurate prediction. The accuracy of our best RF classification model obtained including both surface and local water features is 0.89 ± 0.05. A similar accuracy can be achieved including only local water features, suggesting that the important surface properties are largely captured by the local water features. Some important features identified by ML analysis are local icelike structures, water density and polarization profiles perpendicular to the surface, and the two-dimensional lattice match to ice. We expect that this work, with its strong focus on realistic surface models, will serve as a guide to the identification or design of substrates that can promote or discourage ice nucleation.

The limit of macroscopic homogeneous ice nucleation at the nanoscale

Nucleation in small volumes of water has garnered renewed interest due to the relevance of pore condensation and freezing under conditions of low partial pressures of water, such as in the upper troposphere. Molecular simulations can in principle provide insight on this process at the molecular scale that is challenging to achieve experimentally. However, there are discrepancies in the literature as to whether the rate in confined systems is enhanced or suppressed relative to bulk water at the same temperature and pressure. In this study, we investigate the extent to which the size of the critical nucleus and the rate at which it grows in thin films of water are affected by the thickness of the film. Our results suggest that nucleation remains bulk-like in films that are barely large enough accommodate a critical nucleus. This conclusion seems robust to the presence of physical confining boundaries. We also discuss the difficulties in unambiguously determining homogeneous nucleation rates in nanoscale systems, owing to the challenges in defining the volume. Our results suggest any impact on a film's thickness on the rate is largely inconsequential for present day experiments.

Detecting dynamic domains and local fluctuations in complex molecular systems via timelapse neighbors shuffling

It is known that the behavior of many complex systems is controlled by local dynamic rearrangements or fluctuations occurring within them. Complex molecular systems, composed of many molecules interacting with each other in a Brownian storm, make no exception. Despite the rise of machine learning and of sophisticated structural descriptors, detecting local fluctuations and collective transitions in complex dynamic ensembles remains often difficult. Here, we show a machine learning framework based on a descriptor which we name Local Environments and Neighbors Shuffling (LENS), that allows identifying dynamic domains and detecting local fluctuations in a variety of systems in an abstract and efficient way. By tracking how much the microscopic surrounding of each molecular unit changes over time in terms of neighbor individuals, LENS allows characterizing the global (macroscopic) dynamics of molecular systems in phase transition, phases-coexistence, as well as intrinsically characterized by local fluctuations (e.g., defects). Statistical analysis of the LENS time series data extracted from molecular dynamics trajectories of, for example, liquid-like, solid-like, or dynamically diverse complex molecular systems allows tracking in an efficient way the presence of different dynamic domains and of local fluctuations emerging within them. The approach is found robust, versatile, and applicable independently of the features of the system and simply provided that a trajectory containing information on the relative motion of the interacting units is available. We envisage that "such a LENS" will constitute a precious basis for exploring the dynamic complexity of a variety of systems and, given its abstract definition, not necessarily of molecular ones.

Controlling crystallization: what liquid structure and dynamics reveal about crystal nucleation mechanisms

Over recent years, molecular simulations have provided invaluable insights into the microscopic processes governing the initial stages of crystal nucleation and growth. A key aspect that has been observed in many different systems is the formation of precursors in the supercooled liquid that precedes the emergence of crystalline nuclei. The structural and dynamical properties of these precursors determine to a large extent the nucleation probability as well as the formation of specific polymorphs. This novel microscopic view on nucleation mechanisms has further implications for our understanding of the nucleating ability and polymorph selectivity of nucleating agents, as these appear to be strongly linked to their ability in modifying structural and dynamical characteristics of the supercooled liquid, namely liquid heterogeneity. In this perspective, we highlight recent progress in exploring the connection between liquid heterogeneity and crystallization, including the effects of templates, and the potential impact for controlling crystallization processes. This article is part of a discussion meeting issue 'Supercomputing simulations of advanced materials'.

Characterizing Surface Ice-Philicity Using Molecular Simulations and Enhanced Sampling

The formation of ice, which plays an important role in diverse contexts ranging from cryopreservation to atmospheric science, is often mediated by solid surfaces. Although surfaces that interact favorably with ice (relative to liquid water) can facilitate ice formation by lowering nucleation barriers, the molecular characteristics that confer icephilicity to a surface are complex and incompletely understood. To address this challenge, here we introduce a robust and computationally efficient method for characterizing surface ice-philicity that combines molecular simulations and enhanced sampling techniques to quantify the free energetic cost of increasing surface-ice contact at the expense of surface-water contact. Using this method to characterize the ice-philicity of a family of model surfaces that are lattice matched with ice but vary in their polarity, we find that the nonpolar surfaces are moderately ice-phobic, whereas the polar surfaces are highly ice-philic. In contrast, for surfaces that display no complementarity to the ice lattice, we find that ice-philicity is independent of surface polarity and that both nonpolar and polar surfaces are moderately ice-phobic. Our work thus provides a prescription for quantitatively characterizing surface ice-philicity and sheds light on how ice-philicity is influenced by lattice matching and polarity.

Melting domain size and recrystallization dynamics of ice revealed by time-resolved x-ray scattering

The phase transition between water and ice is ubiquitous and one of the most important phenomena in nature. Here, we performed time-resolved x-ray scattering experiments capturing the melting and recrystallization dynamics of ice. The ultrafast heating of ice I is induced by an IR laser pulse and probed with an intense x-ray pulse which provided us with direct structural information on different length scales. From the wide-angle x-ray scattering (WAXS) patterns, the molten fraction, as well as the corresponding temperature at each delay, were determined. The small-angle x-ray scattering (SAXS) patterns, together with the information extracted from the WAXS analysis, provided the time-dependent change of the size and the number of liquid domains. The results show partial melting (~13%) and superheating of ice occurring at around 20 ns. After 100 ns, the average size of the liquid domains grows from about 2.5 nm to 4.5 nm by the coalescence of approximately six adjacent domains. Subsequently, we capture the recrystallization of the liquid domains, which occurs on microsecond timescales due to the cooling by heat dissipation and results to a decrease of the average liquid domain size.

Are Neural Network Potentials Trained on Liquid States Transferable to Crystal Nucleation? A Test on Ice Nucleation in the mW Water Model

Neural network potentials (NNPs) are increasingly being used to study processes that happen on long time scales. A typical example is crystal nucleation, which rate is controlled by the occurrence of a rare fluctuation, i.e., the appearance of the critical nucleus. Because the properties of this nucleus are far from those of the bulk crystal, it is yet unclear whether NN potentials trained on equilibrium liquid states can accurately describe nucleation processes. So far, nucleation studies on NNPs have been limited to ab initio models whose nucleation properties are unknown, preventing an accurate comparison. Here we train a NN potential on the mW model of water─a classical three-body potential whose nucleation time scale is accessible in standard simulations. We show that a NNP trained only on a small number of liquid state points can reproduce with great accuracy the nucleation rates and free energy barriers of the original model, computed from both spontaneous and biased trajectories, strongly supporting the use of NNPs to study nucleation events.

A neural network potential with self-trained atomic fingerprints: A test with the mW water potential

We present a neural network (NN) potential based on a new set of atomic fingerprints built upon two- and three-body contributions that probe distances and local orientational order, respectively. Compared with the existing NN potentials, the atomic fingerprints depend on a small set of tunable parameters that are trained together with the NN weights. In addition to simplifying the selection of the atomic fingerprints, this strategy can also considerably increase the overall accuracy of the network representation. To tackle the simultaneous training of the atomic fingerprint parameters and NN weights, we adopt an annealing protocol that progressively cycles the learning rate, significantly improving the accuracy of the NN potential. We test the performance of the network potential against the mW model of water, which is a classical three-body potential that well captures the anomalies of the liquid phase. Trained on just three state points, the NN potential is able to reproduce the mW model in a very wide range of densities and temperatures, from negative pressures to several GPa, capturing the transition from an open random tetrahedral network to a dense interpenetrated network. The NN potential also reproduces very well properties for which it was not explicitly trained, such as dynamical properties and the structure of the stable crystalline phases of mW.

Phase diagrams-Why they matter and how to predict them

Understanding the thermodynamic stability and metastability of materials can help us to, for example, gauge whether crystalline polymorphs in pharmaceutical formulations are likely to be durable. It can also help us to design experimental routes to novel phases with potentially interesting properties. In this Perspective, we provide an overview of how thermodynamic phase behavior can be quantified both in computer simulations and machine-learning approaches to determine phase diagrams, as well as combinations of the two. We review the basic workflow of free-energy computations for condensed phases, including some practical implementation advice, ranging from the Frenkel-Ladd approach to thermodynamic integration and to direct-coexistence simulations. We illustrate the applications of such methods on a range of systems from materials chemistry to biological phase separation. Finally, we outline some challenges, questions, and practical applications of phase-diagram determination which we believe are likely to be possible to address in the near future using such state-of-the-art free-energy calculations, which may provide fundamental insight into separation processes using multicomponent solvents.

Nanoporous ices: an emerging class in the water/ice family

The history of scientific research on diverse ice structures dates back to more than a century. To date, 20 three-dimensional crystalline ice phases (ice I-ice XX) have been identified in the laboratory, among which ice XVI and ice XVII belong to a class of low-density nanoporous ices. Nanoporous ices can also be viewed as a special class of porous materials or water ice, as they possess a relatively high fraction of nano-cavities and/or nano-channels built into the hydrogen-bonded water framework. As such, like the prototypical class of porous materials (e.g., MOFs and COFs), nanoporous ices can be named as water oxygen-vertex frameworks (WOFs). Because of their large surface-to-volume ratio, WOFs may be potential media for gas storage, gas purification and separation. They may be applied to the biomedical field owing to their excellent biocompatibility. The field of porous ices is still emerging, as many porous ice structures that are predicted to be stable by computer simulations require future experimental confirmation. For future theoretical/computational studies, as the machine-learning method becomes an increasingly popular research tool in the material science and chemical science fields, more reliable porous ice structures and phase diagrams will be predicted with the development of more accurate machine-learning force fields.