Silicon Liquid Structure and Crystal Nucleation from Ab Initio Deep Metadynamics

Physics-Informed Active Learning for Accelerating Quantum Chemical Simulations

Quantum chemical simulations can be greatly accelerated by constructing machine learning potentials, which is often done using active learning (AL). The usefulness of the constructed potentials is often limited by the high effort required and their insufficient robustness in the simulations. Here, we introduce the end-to-end AL for constructing robust data-efficient potentials with affordable investment of time and resources and minimum human interference. Our AL protocol is based on the physics-informed sampling of training points, automatic selection of initial data, uncertainty quantification, and convergence monitoring. The versatility of this protocol is shown in our implementation of quasi-classical molecular dynamics for simulating vibrational spectra, conformer search of a key biochemical molecule, and time-resolved mechanism of the Diels-Alder reaction. These investigations took us days instead of weeks of pure quantum chemical calculations on a high-performance computing cluster.

Enhanced deep potential model for fast and accurate molecular dynamics: application to the hydrated electron

In molecular simulations, neural network force fields aim at achieving ab initio accuracy with reduced computational cost. This work introduces enhancements to the Deep Potential network architecture, integrating a message-passing framework and a new lightweight implementation with various improvements. Our model achieves accuracy on par with leading machine learning force fields and offers significant speed advantages, making it well-suited for large-scale, accuracy-sensitive systems. We also introduce a new iterative model for Wannier center prediction, allowing us to keep track of electron positions in simulations of general insulating systems. We apply our model to study the solvated electron in bulk water, an ostensibly simple system that is actually quite challenging to represent with neural networks. Our trained model is not only accurate, but can also transfer to larger systems. Our simulation confirms the cavity model, where the electron's localized state is observed to be stable. Through an extensive run, we accurately determine various structural and dynamical properties of the solvated electron.

Improving Molecular-Dynamics Simulations for Solid-Liquid Interfaces with Machine-Learning Interatomic Potentials

Emerging developments in artificial intelligence have opened infinite possibilities for material simulation. Depending on the powerful fitting of machine learning algorithms to first-principles data, machine learning interatomic potentials (MLIPs) can effectively balance the accuracy and efficiency problems in molecular dynamics (MD) simulations, serving as powerful tools in various complex physicochemical systems. Consequently, this brings unprecedented enthusiasm for researchers to apply such novel technology in multiple fields to revisit the major scientific problems that have remained controversial owing to the limitations of previous computational methods. Herein, we introduce the evolution of MLIPs, provide valuable application examples for solid-liquid interfaces, and present current challenges. Driven by solving multitudinous difficulties in terms of the accuracy, efficiency, and versatility of MLIPs, this booming technique, combined with molecular simulation methods, will provide an underlying and valuable understanding of interdisciplinary scientific challenges, including materials, physics, and chemistry.

Derivative learning of tensorial quantities-Predicting finite temperature infrared spectra from first principles

We develop a strategy that integrates machine learning and first-principles calculations to achieve technically accurate predictions of infrared spectra. In particular, the methodology allows one to predict infrared spectra for complex systems at finite temperatures. The method's effectiveness is demonstrated in challenging scenarios, such as the analysis of water and the organic-inorganic halide perovskite MAPbI3, where our results consistently align with experimental data. A distinctive feature of the methodology is the incorporation of derivative learning, which proves indispensable for obtaining accurate polarization data in bulk materials and facilitates the training of a machine learning surrogate model of the polarization adapted to rotational and translational symmetries. We achieve polarization prediction accuracies of about 1% for the water dimer by training only on the predicted Born effective charges.

Deciphering diffuse scattering with machine learning and the equivariant foundation model: the case of molten FeO

Bridging the gap between diffuse x-ray or neutron scattering measurements and predicted structures derived from atom-atom pair potentials in disordered materials, has been a longstanding challenge in condensed matter physics. This perspective gives a brief overview of the traditional approaches employed over the past several decades. Namely, the use of approximate interatomic pair potentials that relate three-dimensional structural models to the measured structure factor and its' associated pair distribution function. The use of machine learned interatomic potentials has grown in the past few years, and has been particularly successful in the cases of ionic and oxide systems. Recent advances in large scale sampling, along with a direct integration of scattering measurements into the model development, has provided improved agreement between experiments and large-scale models calculated with quantum mechanical accuracy. However, details of local polyhedral bonding and connectivity in meta-stable disordered systems still require improvement. Here we leverage MACE-MP-0; a newly introduced equivariant foundation model and validate the results against high-quality experimental scattering data for the case of molten iron(II) oxide (FeO). These preliminary results suggest that the emerging foundation model has the potential to surpass the traditional limitations of classical interatomic potentials.

A reactive neural network framework for water-loaded acidic zeolites

Under operating conditions, the dynamics of water and ions confined within protonic aluminosilicate zeolite micropores are responsible for many of their properties, including hydrothermal stability, acidity and catalytic activity. However, due to high computational cost, operando studies of acidic zeolites are currently rare and limited to specific cases and simplified models. In this work, we have developed a reactive neural network potential (NNP) attempting to cover the entire class of acidic zeolites, including the full range of experimentally relevant water concentrations and Si/Al ratios. This NNP has the potential to dramatically improve sampling, retaining the (meta)GGA DFT level accuracy, with the capacity for discovery of new chemistry, such as collective defect formation mechanisms at the zeolite surface. Furthermore, we exemplify how the NNP can be used as a basis for further extensions/improvements which include data-efficient adoption of higher-level (hybrid) references via Δ-learning and the acceleration of rare event sampling via automatic construction of collective variables. These developments represent a significant step towards accurate simulations of realistic catalysts under operando conditions.

Modelling atomic and nanoscale structure in the silicon-oxygen system through active machine learning

Silicon-oxygen compounds are among the most important ones in the natural sciences, occurring as building blocks in minerals and being used in semiconductors and catalysis. Beyond the well-known silicon dioxide, there are phases with different stoichiometric composition and nanostructured composites. One of the key challenges in understanding the Si-O system is therefore to accurately account for its nanoscale heterogeneity beyond the length scale of individual atoms. Here we show that a unified computational description of the full Si-O system is indeed possible, based on atomistic machine learning coupled to an active-learning workflow. We showcase applications to very-high-pressure silica, to surfaces and aerogels, and to the structure of amorphous silicon monoxide. In a wider context, our work illustrates how structural complexity in functional materials beyond the atomic and few-nanometre length scales can be captured with active machine learning.

Structure and polymerization of liquid sulfur across the λ-transition

The anomalous λ-transition of liquid sulfur, which is supposed to be related to the transformation of eight-membered sulfur rings into long polymeric chains, has attracted considerable attention. However, a detailed description of the underlying dynamical polymerization process is still missing. Here, we study the structures and the mechanism of the polymerization processes of liquid sulfur across the λ-transition as well as its reverse process of formation of the rings. We do so by performing ab initio-quality molecular dynamics simulations thanks to a combination of machine learning potentials and state-of-the-art enhanced sampling techniques. With our approach, we obtain structural results that are in good agreement with the experiments and we report precious dynamical insights into the mechanisms involved in the process.

Divergence among Local Structure, Dynamics, and Nucleation Outcome in Heterogeneous Nucleation of Close-Packed Crystals

Heterogeneous crystal nucleation is the dominant mechanism of crystallization in most systems, yet its underlying physics remains an enigma. While emergent interfacial crystalline order precedes heterogeneous nucleation, its importance in the nucleation mechanism is unclear. Here, we use path sampling simulations of two model systems to demonstrate that crystalline order in its traditional sense is not predictive of the outcome of the heterogeneous nucleation of close-packed crystals. Consequently, structure-based collective variables (CVs) that reliably describe homogeneous nucleation can be poor descriptors of heterogeneous nucleation. This divergence between structure and nucleation outcome is accompanied by an intriguing dynamical anomaly, wherein low-coordinated crystalline particles outpace their liquid-like counterparts. We use committor analysis, high-throughput screening, and machine learning to devise CV optimization strategies and present suitable structural heuristics within the metastable fluid for CV prescreening. Employing such optimized CVs is pivotal for properly characterizing the mechanism of heterogeneous nucleation in metallic and colloidal systems.

Temperature-Dependent Paracrystalline Nucleation in Atomically Disordered Diamonds

Atomically disordered diamonds with medium-range order realized in recent experiments extend our knowledge of atomic disorder in materials. However, the current understanding of amorphous carbons cannot answer why paracrystalline diamond (p-D) can be formed inherently different from other tetrahedral amorphous carbons (ta-Cs), and the emergence of p-D seems to be easily hindered by inappropriate temperatures. Herein, we performed atomistic-based simulations to shed light on temperature-dependent paracrystalline nucleation in atomically disordered diamonds. Using metadynamics and two carefully designed collective variables, reversible phase transitions among different ta-Cs can be presented under different temperatures, evidenced by corresponding local minima on the free energy surface and reaction path along the free energy gradient. We found that p-D is preferred in a narrow range of temperatures, which is comparable to real experimental temperatures under the Arrhenius framework. The insights and related methods should open up a perspective for investigating other amorphous carbons.

The role of dynamics in heterogeneous catalysis: Surface diffusivity and N2 decomposition on Fe(111)

Dynamics has long been recognized to play an important role in heterogeneous catalytic processes. However, until recently, it has been impossible to study their dynamical behavior at industry-relevant temperatures. Using a combination of machine learning potentials and advanced simulation techniques, we investigate the cleavage of the N[Formula: see text] triple bond on the Fe(111) surface. We find that at low temperatures our results agree with the well-established picture. However, if we increase the temperature to reach operando conditions, the surface undergoes a global dynamical change and the step structure of the Fe(111) surface is destabilized. The catalytic sites, traditionally associated with this surface, appear and disappear continuously. Our simulations illuminate the danger of extrapolating low-temperature results to operando conditions and indicate that the catalytic activity can only be inferred from calculations that take dynamics fully into account. More than that, they show that it is the transition to this highly fluctuating interfacial environment that drives the catalytic process.

Double-Shock Compression Pathways from Diamond to BC8 Carbon

Carbon is one of the most important elements for both industrial applications and fundamental research, including life, physics, chemistry, materials, and even planetary science. Although theoretical predictions on the transition from diamond to the BC8 (Ia3[over ¯]) carbon were made more than thirty years ago, after tremendous experimental efforts, direct evidence for the existence of BC8 carbon is still lacking. In this study, a machine learning potential was developed for high-pressure carbon fitted from first-principles calculations, which exhibited great capabilities in modeling the melting and Hugoniot line. Using the molecular dynamics based on this machine learning potential, we designed a thermodynamic pathway that is achievable for the double shock compression experiment to obtain the elusive BC8 carbon. Diamond was compressed up to 584 GPa after the first shock at 20.5  km/s. Subsequently, in the second shock compression at 24.8 or 25.0  km/s, diamond was compressed to a supercooled liquid and then solidified to BC8 in around 1 ns. Furthermore, the critical nucleus size and nucleation rate of BC8 were calculated, which are crucial for nano-second x-ray diffraction measurements to observe BC8 carbon during shock compressions. The key to obtaining BC8 carbon lies in the formation of liquid at a sufficient supercooling. Our work provides a feasible pathway by which the long-sought BC8 phase of carbon can be reached in experiments.

Optimizing molecular potential models by imposing kinetic constraints with path reweighting

Empirical force fields employed in molecular dynamics simulations of complex systems are often optimized to reproduce experimentally determined structural and thermodynamic properties. In contrast, experimental knowledge about the interconversion rates between metastable states in such systems is hardly ever incorporated in a force field due to a lack of an efficient approach. Here, we introduce such a framework based on the relationship between dynamical observables, such as rate constants, and the underlying molecular model parameters using the statistical mechanics of trajectories. Given a prior ensemble of molecular dynamics trajectories produced with imperfect force field parameters, the approach allows for the optimal adaption of these parameters such that the imposed constraint of equally predicted and experimental rate constant is obeyed. To do so, the method combines the continuum path ensemble maximum caliber approach with path reweighting methods for stochastic dynamics. When multiple solutions are found, the method selects automatically the combination that corresponds to the smallest perturbation of the entire path ensemble, as required by the maximum entropy principle. To show the validity of the approach, we illustrate the method on simple test systems undergoing rare event dynamics. Next to simple 2D potentials, we explore particle models representing molecular isomerization reactions and protein-ligand unbinding. Besides optimal interaction parameters, the methodology gives physical insights into what parts of the model are most sensitive to the kinetics. We discuss the generality and broad implications of the methodology.

DeePMD-kit v2: A software package for deep potential models

DeePMD-kit is a powerful open-source software package that facilitates molecular dynamics simulations using machine learning potentials known as Deep Potential (DP) models. This package, which was released in 2017, has been widely used in the fields of physics, chemistry, biology, and material science for studying atomistic systems. The current version of DeePMD-kit offers numerous advanced features, such as DeepPot-SE, attention-based and hybrid descriptors, the ability to fit tensile properties, type embedding, model deviation, DP-range correction, DP long range, graphics processing unit support for customized operators, model compression, non-von Neumann molecular dynamics, and improved usability, including documentation, compiled binary packages, graphical user interfaces, and application programming interfaces. This article presents an overview of the current major version of the DeePMD-kit package, highlighting its features and technical details. Additionally, this article presents a comprehensive procedure for conducting molecular dynamics as a representative application, benchmarks the accuracy and efficiency of different models, and discusses ongoing developments.

Anisotropic Collective Variables with Machine Learning Potential for Ab Initio Crystallization of Complex Ceramics

Enhanced sampling molecular dynamics (MD) simulations have been extensively used in the phase transition study of simple crystalline materials, such as aluminum, silica, and ice. However, MD simulation of the crystallization process for complex crystalline materials still faces a formidable challenge due to their multicomponent induced multiphase problem. Here, we realize the ab initio accuracy MD crystallization simulations of complex ceramics by using anisotropic collective variables (CVs) and machine learning (ML) potential. The anisotropic X-ray diffraction intensity CVs provide precise identification of complex crystal structures with detailed crystallography information, while the ML potential makes it feasible to further perform enhanced sampling simulations with ab initio accuracy. We verify the universality and accuracy of this method through complex ceramics with three kinds of representative structures, i.e., Ti3SiC2 for the MAX structure, zircon for the mineral structure, and lead zirconate titanate for the perovskite structure. It demonstrates exceptional efficiency and ab initio quality in achieving crystallization and generating free energy surfaces of all these ceramics, facilitating the analysis and design of complex crystalline materials.

Temperature-pressure phase diagram of confined monolayer water/ice at first-principles accuracy with a machine-learning force field

Understanding the phase behaviour of nanoconfined water films is of fundamental importance in broad fields of science and engineering. However, the phase behaviour of the thinnest water film - monolayer water - is still incompletely known. Here, we developed a machine-learning force field (MLFF) at first-principles accuracy to determine the phase diagram of monolayer water/ice in nanoconfinement with hydrophobic walls. We observed the spontaneous formation of two previously unreported high-density ices, namely, zigzag quasi-bilayer ice (ZZ-qBI) and branched-zigzag quasi-bilayer ice (bZZ-qBI). Unlike conventional bilayer ices, few inter-layer hydrogen bonds were observed in both quasi-bilayer ices. Notably, the bZZ-qBI entails a unique hydrogen-bonding network that consists of two distinctive types of hydrogen bonds. Moreover, we identified, for the first time, the stable region for the lowest-density [Formula: see text] monolayer ice (LD-48MI) at negative pressures (<-0.3 GPa). Overall, the MLFF enables large-scale first-principle-level molecular dynamics (MD) simulations of the spontaneous transition from the liquid water to a plethora of monolayer ices, including hexagonal, pentagonal, square, zigzag (ZZMI), and hexatic monolayer ices. These findings will enrich our understanding of the phase behaviour of the nanoconfined water/ices and provide a guide for future experimental realization of the 2D ices.

High-Accuracy Neural Network Interatomic Potential for Silicon Nitride

In the field of machine learning (ML) and data science, it is meaningful to use the advantages of ML to create reliable interatomic potentials. Deep potential molecular dynamics (DEEPMD) are one of the most useful methods to create interatomic potentials. Among ceramic materials, amorphous silicon nitride (SiN_x) features good electrical insulation, abrasion resistance, and mechanical strength, which is widely applied in industries. In our work, a neural network potential (NNP) for SiN_x was created based on DEEPMD, and the NNP is confirmed to be applicable to the SiN_x model. The tensile tests were simulated to compare the mechanical properties of SiN_x with different compositions based on the molecular dynamic method coupled with NNP. Among these SiN_x, Si₃N₄ has the largest elastic modulus (E) and yield stress (σ_s), showing the desired mechanical strength owing to the largest coordination numbers (CN) and radial distribution function (RDF). The RDFs and CNs decrease with the increase of x; meanwhile, E and σ_s of SiN_x decrease when the proportion of Si increases. It can be concluded that the ratio of nitrogen to silicon can reflect the RDFs and CNs in micro level and macro mechanical properties of SiN_x to a large extent.

Collective Variables for Crystallization Simulations-from Early Developments to Recent Advances

Crystallization is an important physicochemical process which has relevance in material science, biology, and the environment. Decades of experimental and theoretical efforts have been made to understand this fundamental symmetry-breaking transition. While experiments provide equilibrium structures and shapes of crystals, they are limited to unraveling how molecules aggregate to form crystal nuclei that subsequently transform into bulk crystals. Computer simulations, mainly molecular dynamics (MD), can provide such microscopic details during the early stage of a crystallization event. Crystallization is a rare event that takes place in time scales much longer than a typical equilibrium MD simulation can sample. This inadequate sampling of the MD method can be easily circumvented by the use of enhanced sampling (ES) simulations. In most of the ES methods, the fluctuations of a system's slow degrees of freedom, called collective variables (CVs), are enhanced by applying a bias potential. This transforms the system from one state to the other within a short time scale. The most crucial part of such CV-based ES methods is to find suitable CVs, which often needs intuition and several trial-and-error optimization steps. Over the years, a plethora of CVs has been developed and applied in the study of crystallization. In this review, we provide a brief overview of CVs that have been developed and used in ES simulations to study crystallization from melt or solution. These CVs can be categorized mainly into four types: (i) spherical particle-based, (ii) molecular template-based, (iii) physical property-based, and (iv) CVs obtained from dimensionality reduction techniques. We present the context-based evolution of CVs, discuss the current challenges, and propose future directions to further develop effective CVs for the study of crystallization of complex systems.

Solvation structures of calcium and magnesium ions in water with the presence of hydroxide: a study by deep potential molecular dynamics

The solvation structures of calcium (Ca²⁺) and magnesium (Mg²⁺) ions with the presence of hydroxide (OH^-) ion in water are essential for understanding their roles in biological and chemical processes but have not been fully explored. Ab initio molecular dynamics (AIMD) is an important tool to address this issue, but two challenges exist. First, an accurate description of OH^- from AIMD needs an appropriate exchange-correlation functional. Second, a long trajectory is needed to reach an equilibrium state for the Ca²⁺-OH^- and Mg²⁺-OH^- ion pairs in aqueous solutions. Herein, we adopt a deep potential molecular dynamics (DPMD) method to simulate 1 ns trajectories for the Ca²⁺-OH^- and Mg²⁺-OH^- ion pairs in water; the DPMD method provides efficient machine-learning-based models that have the accuracy of the SCAN exchange-correlation functional within the framework of density functional theory. The solvation structures of the cations and the OH^- in terms of three different species have been systematically investigated. On the one hand, we find that OH^- have more significant effects on the solvation structure of Ca²⁺ than that of Mg²⁺. We observe that the OH^- substantially affects the orientation angles of water molecules surrounding the cation. Through the time correlation functions, we conclude that the water molecules in the first solvation shell of Ca²⁺ change their preferred orientation faster than those of Mg²⁺. On the other hand, with the presence of the cation in the first solvation shell of OH^-, we find that the hydrogen bonds of OH^- are severely altered, and the adjacent water molecules of OH^- are squeezed. The two cations have substantially different effects on the solvation structure of OH^-. Our work provides new insight into the solvation structures of Ca²⁺ and Mg²⁺ in water with the presence of OH^-.

Deep coarse-grained potentials via relative entropy minimization

Neural network (NN) potentials are a natural choice for coarse-grained (CG) models. Their many-body capacity allows highly accurate approximations of the potential of mean force, promising CG simulations of unprecedented accuracy. CG NN potentials trained bottom-up via force matching (FM), however, suffer from finite data effects: They rely on prior potentials for physically sound predictions outside the training data domain, and the corresponding free energy surface is sensitive to errors in the transition regions. The standard alternative to FM for classical potentials is relative entropy (RE) minimization, which has not yet been applied to NN potentials. In this work, we demonstrate, for benchmark problems of liquid water and alanine dipeptide, that RE training is more data efficient, due to accessing the CG distribution during training, resulting in improved free energy surfaces and reduced sensitivity to prior potentials. In addition, RE learns to correct time integration errors, allowing larger time steps in CG molecular dynamics simulation, while maintaining accuracy. Thus, our findings support the use of training objectives beyond FM, as a promising direction for improving CG NN potential's accuracy and reliability.

Stochastic Resetting for Enhanced Sampling

We present a method for enhanced sampling of molecular dynamics simulations using stochastic resetting. Various phenomena, ranging from crystal nucleation to protein folding, occur on time scales that are unreachable in standard simulations. They are often characterized by broad transition time distributions, in which extremely slow events have a non-negligible probability. Stochastic resetting, i.e., restarting simulations at random times, was recently shown to significantly expedite processes that follow such distributions. Here, we employ resetting for enhanced sampling of molecular simulations for the first time. We show that it accelerates long time scale processes by up to an order of magnitude in examples ranging from simple models to a molecular system. Most importantly, we recover the mean transition time without resetting, which is typically too long to be sampled directly, from accelerated simulations at a single restart rate. Stochastic resetting can be used as a standalone method or combined with other sampling algorithms to further accelerate simulations.

Machine learning interatomic potentials for aluminium: application to solidification phenomena

In studying solidification process by simulations on the atomic scale, the modeling of crystal nucleation or amorphization requires the construction of interatomic interactions that are able to reproduce the properties of both the solid and the liquid states. Taking into account rare nucleation events or structural relaxation under deep undercooling conditions requires much larger length scales and longer time scales than those achievable byab initiomolecular dynamics (AIMD). This problem is addressed by means of classical molecular dynamics simulations using a well established high dimensional neural network potential trained on a set of configurations generated by AIMD relevant for solidification phenomena. Our dataset contains various crystalline structures and liquid states at different pressures, including their time fluctuations in a wide range of temperatures. Applied to elemental aluminium, the resulting potential is shown to be efficient to reproduce the basic structural, dynamics and thermodynamic quantities in the liquid and undercooled states. Early stages of crystallization are further investigated on a much larger scale with one million atoms, allowing us to unravel features of the homogeneous nucleation mechanisms in the fcc phase at ambient pressure as well as in the bcc phase at high pressure with unprecedented accuracy close to theab initioone. In both cases, a single step nucleation process is observed.

Size-Dependent Nucleation in Crystal Phase Transition from Machine Learning Metadynamics

In this Letter, we present a framework that combines machine learning potential (MLP) and metadynamics to investigate solid-solid phase transition. Based on the spectral descriptors and neural networks regression, we develop a scalable MLP model to warrant an accurate interpolation of the energy surface where two phases coexist. Applying it to the simulation of B4-B1 phase transition of GaN under 50 GPa with different model sizes, we observe sequential change of the phase transition mechanism from collective modes to nucleation and growths. When the size is at or below 128 000 atoms, the nucleation and growth appear to follow a preferred direction. At larger sizes, the nuclei occur at multiple sites simultaneously and grow to microstructures by passing the critical size. The observed change of the atomistic mechanism manifests the importance of statistical sampling with large system size in phase transition modeling.

Capabilities and limits of autoencoders for extracting collective variables in atomistic materials science

Free energy calculations in materials science are routinely hindered by the need to provide reaction coordinates that can meaningfully partition atomic configuration space, a prerequisite for most enhanced sampling approaches. Recent studies on molecular systems have highlighted the possibility of constructing appropriate collective variables directly from atomic motions through deep learning techniques. Here we extend this class of approaches to condensed matter problems, for which we encode the finite temperature collective variable by an iterative procedure starting from 0 K features of the energy landscape i.e. activation events or migration mechanisms given by a minimum - saddle point - minimum sequence. We employ the autoencoder neural networks in order to build a scalar collective variable for use with the adaptive biasing force method. Particular attention is given to design choices required for application to crystalline systems with defects, including the filtering of thermal motions which otherwise dominate the autoencoder input. The machine-learning workflow is tested on body-centered cubic iron and its common defects, such as small vacancy or self-interstitial clusters and screw dislocations. For localized defects, excellent collective variables as well as derivatives, necessary for free energy sampling, are systematically obtained. However, the approach has a limited accuracy when dealing with reaction coordinates that include atomic displacements of a magnitude comparable to thermal motions, e.g. the ones produced by the long-range elastic field of dislocations. We then combine the extraction of collective variables by autoencoders with an adaptive biasing force free energy method based on Bayesian inference. Using a vacancy migration as an example, we demonstrate the performance of coupling these two approaches for simultaneous discovery of reaction coordinates and free energy sampling in systems with localized defects.

On Sampling Minimum Energy Path

Sampling the minimum energy path (MEP) between two minima of a system is often hindered by the presence of an energy barrier separating the two metastable states. As a consequence, direct sampling based on molecular dynamics or Markov Chain Monte Carlo methods becomes inefficient, the crossing of the energy barrier being associated to a rare event. Augmented sampling methods based on the definition of collective variables or reaction coordinates allow us to circumvent this limitation at the price of an arbitrary choice of the dimensionality reduction algorithm. We couple the statistical sampling techniques, namely, metadynamics and invertible neural networks, with autoencoders so as to gradually learn the MEP and the collective variable at the same time. Learning is achieved through a succession of two steps: statistical sampling of the most probable path between the two minima and redefinition of the collective variable from the updated data points. The prototypical Mueller potential with nearly orthogonal minima is employed to demonstrate the ability of such coupling to unravel a complex MEP.

Homogeneous ice nucleation in an ab initio machine-learning model of water

Molecular simulations have provided valuable insight into the microscopic mechanisms underlying homogeneous ice nucleation. While empirical models have been used extensively to study this phenomenon, simulations based on first-principles calculations have so far proven prohibitively expensive. Here, we circumvent this difficulty by using an efficient machine-learning model trained on density-functional theory energies and forces. We compute nucleation rates at atmospheric pressure, over a broad range of supercoolings, using the seeding technique and systems of up to hundreds of thousands of atoms simulated with ab initio accuracy. The key quantity provided by the seeding technique is the size of the critical cluster (i.e., a size such that the cluster has equal probabilities of growing or melting at the given supersaturation), which is used together with the equations of classical nucleation theory to compute nucleation rates. We find that nucleation rates for our model at moderate supercoolings are in good agreement with experimental measurements within the error of our calculation. We also study the impact of properties such as the thermodynamic driving force, interfacial free energy, and stacking disorder on the calculated rates.

DP Compress: A Model Compression Scheme for Generating Efficient Deep Potential Models

Machine-learning-based interatomic potential energy surface (PES) models are revolutionizing the field of molecular modeling. However, although much faster than electronic structure schemes, these models suffer from costly computations via deep neural networks to predict the energy and atomic forces, resulting in lower running efficiency as compared to the typical empirical force fields. Herein, we report a model compression scheme for boosting the performance of the Deep Potential (DP) model, a deep learning-based PES model. This scheme, we call DP Compress, is an efficient postprocessing step after the training of DP models (DP Train). DP Compress combines several DP-specific compression techniques, which typically speed up DP-based molecular dynamics simulations by an order of magnitude faster and consume an order of magnitude less memory. We demonstrate that DP Compress is sufficiently accurate by testing a variety of physical properties of Cu, H₂O, and Al-Cu-Mg systems. DP Compress applies to both CPU and GPU machines and is publicly available online.

Theoretical methods for structural phase transitions in elemental solids at extreme conditions: statics and dynamics

In recent years, theoretical studies have moved from a traditionally supporting role to a more proactive role in the research of phase transitions at high pressures. In many cases, theoretical prediction leads the experimental exploration. This is largely owing to the rapid progress of computer power and theoretical methods, particularly the structure prediction methods tailored for high-pressure applications. This review introduces commonly used structure searching techniques based on static and dynamic approaches, their applicability in studying phase transitions at high pressure, and new developments made toward predicting complex crystalline phases. Successful landmark studies for each method are discussed, with an emphasis on elemental solids and their behaviors under high pressure. The review concludes with a perspective on outstanding challenges and opportunities in the field.

Accelerated Deep Learning Dynamics for Atomic Layer Deposition of Al(Me)3 and Water on OH/Si(111)

Knowledge of the detailed mechanism behind the atomic layer deposition (ALD) can greatly facilitate the optimization of the manufacturing process. Computational modeling can potentially foster the understanding; however, the presently available capabilities of the accurate ab initio computational techniques preclude their application to modeling surface processes occurring on a long time scale, such as ALD. Although the situation can be greatly improved using machine learning (ML), this technique requires an enormous amount of data for training datasets. Here, we propose an iterative protocol for optimizing ML training datasets and apply ML-assisted ab initio calculations to model surface reactions occurring during the Al(Me)₃/H₂O ALD process on the OH-terminated Si (111) surface. The protocol uses a recently developed low-dimensional projection technique (TDUS), greatly reducing the amount of information required to achieve high accuracy (ca. 1 kcal/mol or less) of the developed ML models. The resulting free energy landscapes reveal fine details of various aspects of the target ALD process, such as the surface proton transfer, zwitterionic surface configurations, elimination-addition/addition-elimination, and S_N2 reactions as well as the role of the surface entropic and temperature effects. Simulations of adsorption dynamics predict that the maximum physisorption rate of ca. 70% is achieved at the incidence velocity u_rms of the reactants in the range of 15-20 Å/ps. Hence, the proposed protocol furnishes a very effective tool to study complex chemical reaction dynamics at a much reduced computational cost.

Reaction processes at step edges on S-decorated Cu(111) and Ag(111) surfaces: MD analysis utilizing machine learning derived potentials

A variety of complexation, reconstruction, and sulfide formation processes can occur at step edges on the {111} surfaces of coinage metals (M) in the presence of adsorbed S under ultra-high vacuum conditions. Given the cooperative many-atom nature of these reaction processes, Molecular Dynamics (MD) simulation of the associated dynamics is instructive. However, only quite restricted Density Functional Theory (DFT)-level ab initio MD is viable. Thus, for M = Ag and Cu, we instead utilize the DeePMD framework to develop machine-learning derived potentials, retaining near-DFT accuracy for the M–S systems, which should have broad applicability. These potentials are validated by comparison with DFT predictions for various key quantities related to the energetics of S on M(111) surfaces. The potentials are then utilized to perform extensive MD simulations elucidating the above diverse restructuring and reaction processes at step edges. Key observations from MD simulations include the formation of small metal–sulfur complexes, especially MS2; development of a local reconstruction at A-steps featuring an S-decorated {100} motif; and 3D sulfide formation. Additional analysis yields further information on the kinetics for metal–sulfur complex formation, where these complexes can strongly enhance surface mass transport, and on the propensity for sulfide formation.

A deep potential model with long-range electrostatic interactions

Machine learning models for the potential energy of multi-atomic systems, such as the deep potential (DP) model, make molecular simulations with the accuracy of quantum mechanical density functional theory possible at a cost only moderately higher than that of empirical force fields. However, the majority of these models lack explicit long-range interactions and fail to describe properties that derive from the Coulombic tail of the forces. To overcome this limitation, we extend the DP model by approximating the long-range electrostatic interaction between ions (nuclei + core electrons) and valence electrons with that of distributions of spherical Gaussian charges located at ionic and electronic sites. The latter are rigorously defined in terms of the centers of the maximally localized Wannier distributions, whose dependence on the local atomic environment is modeled accurately by a deep neural network. In the DP long-range (DPLR) model, the electrostatic energy of the Gaussian charge system is added to short-range interactions that are represented as in the standard DP model. The resulting potential energy surface is smooth and possesses analytical forces and virial. Missing effects in the standard DP scheme are recovered, improving on accuracy and predictive power. By including long-range electrostatics, DPLR correctly extrapolates to large systems the potential energy surface learned from quantum mechanical calculations on smaller systems. We illustrate the approach with three examples: the potential energy profile of the water dimer, the free energy of interaction of a water molecule with a liquid water slab, and the phonon dispersion curves of the NaCl crystal.

Self-consistent determination of long-range electrostatics in neural network potentials

Machine learning has the potential to revolutionize the field of molecular simulation through the development of efficient and accurate models of interatomic interactions. Neural networks can model interactions with the accuracy of quantum mechanics-based calculations, but with a fraction of the cost, enabling simulations of large systems over long timescales. However, implicit in the construction of neural network potentials is an assumption of locality, wherein atomic arrangements on the nanometer-scale are used to learn interatomic interactions. Because of this assumption, the resulting neural network models cannot describe long-range interactions that play critical roles in dielectric screening and chemical reactivity. Here, we address this issue by introducing the self-consistent field neural network - a general approach for learning the long-range response of molecular systems in neural network potentials that relies on a physically meaningful separation of the interatomic interactions - and demonstrate its utility by modeling liquid water with and without applied fields.

Fast Generation of Machine Learning-Based Force Fields for Adsorption Energies

Adsorption and desorption of molecules are key processes in extraction and purification of biomolecules, engineering of drug carriers, and designing of surface-specific coatings. To understand the adsorption process on the atomic scale, state-of-the-art quantum mechanical and classical simulation methodologies are widely used. However, studying adsorption using a full quantum mechanical treatment is limited to picoseconds simulation timescales, while classical molecular dynamics simulations are limited by the accuracy of the existing force fields. To overcome these challenges, we propose a systematic way to generate flexible, application-specific highly accurate force fields by training artificial neural networks. As a proof of concept, we study the adsorption of the amino acid alanine on graphene and gold (111) surfaces and demonstrate the force field generation methodology in detail. We find that a molecule-specific force field with Lennard-Jones type two-body terms incorporating the 3rd and 7th power of the inverse distances between the atoms of the adsorbent and the surfaces yields optimal results, which is surprisingly different from typical Lennard-Jones potentials used in traditional force fields. Furthermore, we present an efficient and easy-to-train machine learning model that incorporates system-specific three-body (or higher order) interactions that are required, for example, for gold surfaces. Our final machine learning-based force field yields a mean absolute error of less than 4.2 kJ/mol at a speed-up of ∼10⁵ times compared to quantum mechanical calculation, which will have a significant impact on the study of adsorption in different research areas.

Deep learning the slow modes for rare events sampling

The development of enhanced sampling methods has greatly extended the scope of atomistic simulations, allowing long-time phenomena to be studied with accessible computational resources. Many such methods rely on the identification of an appropriate set of collective variables. These are meant to describe the system's modes that most slowly approach equilibrium under the action of the sampling algorithm. Once identified, the equilibration of these modes is accelerated by the enhanced sampling method of choice. An attractive way of determining the collective variables is to relate them to the eigenfunctions and eigenvalues of the transfer operator. Unfortunately, this requires knowing the long-term dynamics of the system beforehand, which is generally not available. However, we have recently shown that it is indeed possible to determine efficient collective variables starting from biased simulations. In this paper, we bring the power of machine learning and the efficiency of the recently developed on the fly probability-enhanced sampling method to bear on this approach. The result is a powerful and robust algorithm that, given an initial enhanced sampling simulation performed with trial collective variables or generalized ensembles, extracts transfer operator eigenfunctions using a neural network ansatz and then accelerates them to promote sampling of rare events. To illustrate the generality of this approach, we apply it to several systems, ranging from the conformational transition of a small molecule to the folding of a miniprotein and the study of materials crystallization.

Learning intermolecular forces at liquid-vapor interfaces

By adopting a perspective informed by contemporary liquid-state theory, we consider how to train an artificial neural network potential to describe inhomogeneous, disordered systems. We find that neural network potentials based on local representations of atomic environments are capable of describing some properties of liquid-vapor interfaces but typically fail for properties that depend on unbalanced long-ranged interactions that build up in the presence of broken translation symmetry. These same interactions cancel in the translationally invariant bulk, allowing local neural network potentials to describe bulk properties correctly. By incorporating explicit models of the slowly varying long-ranged interactions and training neural networks only on the short-ranged components, we can arrive at potentials that robustly recover interfacial properties. We find that local neural network models can sometimes approximate a local molecular field potential to correct for the truncated interactions, but this behavior is variable and hard to learn. Generally, we find that models with explicit electrostatics are easier to train and have higher accuracy. We demonstrate this perspective in a simple model of an asymmetric dipolar fluid, where the exact long-ranged interaction is known, and in an ab initio water model, where it is approximated.

Machine Learning Force Fields

In recent years, the use of machine learning (ML) in computational chemistry has enabled numerous advances previously out of reach due to the computational complexity of traditional electronic-structure methods. One of the most promising applications is the construction of ML-based force fields (FFs), with the aim to narrow the gap between the accuracy of ab initio methods and the efficiency of classical FFs. The key idea is to learn the statistical relation between chemical structure and potential energy without relying on a preconceived notion of fixed chemical bonds or knowledge about the relevant interactions. Such universal ML approximations are in principle only limited by the quality and quantity of the reference data used to train them. This review gives an overview of applications of ML-FFs and the chemical insights that can be obtained from them. The core concepts underlying ML-FFs are described in detail, and a step-by-step guide for constructing and testing them from scratch is given. The text concludes with a discussion of the challenges that remain to be overcome by the next generation of ML-FFs.

Liquid-Liquid Critical Point in Phosphorus

The study of liquid-liquid phase transitions has attracted considerable attention. One interesting example of this phenomenon is phosphorus, for which the existence of a first-order phase transition between a low density insulating molecular phase and a conducting polymeric phase has been experimentally established. In this Letter, we model this transition by an ab initio quality molecular dynamics simulation and explore a large portion of the liquid section of the phase diagram. We draw the liquid-liquid coexistence curve and determine that it terminates into a second-order critical point. Close to the critical point, large coupled structure and electronic structure fluctuations are observed.

The seven deadly sins: When computing crystal nucleation rates, the devil is in the details

The formation of crystals has proven to be one of the most challenging phase transformations to quantitatively model-let alone to actually understand-be it by means of the latest experimental technique or the full arsenal of enhanced sampling approaches at our disposal. One of the most crucial quantities involved with the crystallization process is the nucleation rate, a single elusive number that is supposed to quantify the average probability for a nucleus of critical size to occur within a certain volume and time span. A substantial amount of effort has been devoted to attempt a connection between the crystal nucleation rates computed by means of atomistic simulations and their experimentally measured counterparts. Sadly, this endeavor almost invariably fails to some extent, with the venerable classical nucleation theory typically blamed as the main culprit. Here, we review some of the recent advances in the field, focusing on a number of perhaps more subtle details that are sometimes overlooked when computing nucleation rates. We believe it is important for the community to be aware of the full impact of aspects, such as finite size effects and slow dynamics, that often introduce inconspicuous and yet non-negligible sources of uncertainty into our simulations. In fact, it is key to obtain robust and reproducible trends to be leveraged so as to shed new light on the kinetics of a process, that of crystal nucleation, which is involved into countless practical applications, from the formulation of pharmaceutical drugs to the manufacturing of nano-electronic devices.

Machine learning-accelerated quantum mechanics-based atomistic simulations for industrial applications

Atomistic simulations have become an invaluable tool for industrial applications ranging from the optimization of protein-ligand interactions for drug discovery to the design of new materials for energy applications. Here we review recent advances in the use of machine learning (ML) methods for accelerated simulations based on a quantum mechanical (QM) description of the system. We show how recent progress in ML methods has dramatically extended the applicability range of conventional QM-based simulations, allowing to calculate industrially relevant properties with enhanced accuracy, at reduced computational cost, and for length and time scales that would have otherwise not been accessible. We illustrate the benefits of ML-accelerated atomistic simulations for industrial R&D processes by showcasing relevant applications from two very different areas, drug discovery (pharmaceuticals) and energy materials. Writing from the perspective of both a molecular and a materials modeling scientist, this review aims to provide a unified picture of the impact of ML-accelerated atomistic simulations on the pharmaceutical, chemical, and materials industries and gives an outlook on the exciting opportunities that could emerge in the future.

Neural network potential from bispectrum components: A case study on crystalline silicon

In this article, we present a systematic study on developing machine learning force fields (MLFFs) for crystalline silicon. While the main-stream approach of fitting a MLFF is to use a small and localized training set from molecular dynamics simulations, it is unlikely to cover the global features of the potential energy surface. To remedy this issue, we used randomly generated symmetrical crystal structures to train a more general Si-MLFF. Furthermore, we performed substantial benchmarks among different choices of material descriptors and regression techniques on two different sets of silicon data. Our results show that neural network potential fitting with bispectrum coefficients as descriptors is a feasible method for obtaining accurate and transferable MLFFs.

Spatial Charge Separation as the Origin of Anomalous Stark Effect in Fluorous 2D Hybrid Perovskites

2D hybrid perovskites (2DP) are versatile materials, whose electronic and optical properties can be tuned through the nature of the organic cations (even when those are seemingly electronically inert). Here, it is demonstrated that fluorination of the organic ligands yields glassy 2DP materials featuring long-lived correlated electron-hole pairs. Such states have a marked charge-transfer character, as revealed by the persistent Stark effect in the form of a second derivative in electroabsorption. Modeling shows that electrostatic effects associated with fluorination, combined with the steric hindrance due to the bulky side groups, drive the formation of spatially dislocated charge pairs with reduced recombination rates. This work enriches and broadens the current knowledge of the photophysics of 2DP, which will hopefully guide synthesis efforts toward novel materials with improved functionalities.