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.

New parallelizable schemes for integrating the Dissipative Particle Dynamics with Energy conservation.

This work presents new parallelizable numerical schemes for the integration of dissipative particle dynamics with energy conservation. So far, no numerical scheme introduced in the literature is able to correctly preserve the energy over long times and give rise to small errors on average properties for moderately small time steps, while being straightforwardly parallelizable. We present in this article two new methods, both straightforwardly parallelizable, allowing to correctly preserve the total energy of the system. We illustrate the accuracy and performance of these new schemes both on equilibrium and nonequilibrium parallel simulations.

Characteristics of energy exchange between inter- and intramolecular degrees of freedom in crystalline 1,3,5-triamino-2,4,6-trinitrobenzene (TATB) with implications for coarse-grained simulations of shock waves in polyatomic molecular crystals.

In this report, we characterize the kinetics and dynamics of energy exchange between intramolecular and intermolecular degrees of freedom (DoF) in crystalline 1,3,5-triamino-2,4,6-trinitrobenzene (TATB). All-atom molecular dynamics (MD) simulations are used to obtain predictions for relaxation from certain limiting initial distributions of energy between the intra- and intermolecular DoF. The results are used to parameterize a coarse-grained Dissipative Particle Dynamics at constant Energy (DPDE) model for TATB. Each TATB molecule in the DPDE model is represented as an all-atom, rigid-molecule mesoparticle, with explicit external (molecular translational and rotational) DoF and coarse-grained implicit internal (vibrational) DoF. In addition to conserving linear and angular momentum, the DPDE equations of motion conserve the total system energy provided that particles can exchange energy between their external and internal DoF. The internal temperature of a TATB molecule is calculated using an internal equation of state, which we develop here, and the temperatures of the external and internal DoF are coupled using a fluctuation-dissipation relation. The DPDE force expression requires specification of the input parameter σ that determines the rate at which energy is exchanged between external and internal DoF. We adjusted σ based on the predictions for relaxation processes obtained from MD simulations. The parameterized DPDE model was employed in large-scale simulations of shock compression of TATB. We show that the rate of energy exchange governed by σ can significantly influence the transient behavior of the system behind the shock.

Permutation-invariant distance between atomic configurations.

We present a permutation-invariant distance between atomic configurations, defined through a functional representation of atomic positions. This distance enables us to directly compare different atomic environments with an arbitrary number of particles, without going through a space of reduced dimensionality (i.e., fingerprints) as an intermediate step. Moreover, this distance is naturally invariant through permutations of atoms, avoiding the time consuming associated minimization required by other common criteria (like the root mean square distance). Finally, the invariance through global rotations is accounted for by a minimization procedure in the space of rotations solved by Monte Carlo simulated annealing. A formal framework is also introduced, showing that the distance we propose verifies the property of a metric on the space of atomic configurations. Two examples of applications are proposed. The first one consists in evaluating faithfulness of some fingerprints (or descriptors), i.e., their capacity to represent the structural information of a configuration. The second application concerns structural analysis, where our distance proves to be efficient in discriminating different local structures and even classifying their degree of similarity.

Local density dependent potential for compressible mesoparticles.

This work proposes a coarse grained description of molecular systems based on mesoparticles representing several molecules, where interactions between mesoparticles are modelled by an interparticle potential of molecular type. Since strong non-equilibrium situations over a wide range of pressure and density are targeted, the internal compressibility of the mesoparticles has to be considered. This is done by introducing a dependence of the potential on the local environment of the mesoparticles. To define local densities, we resort to a three-dimensional Voronoi tessellation instead of standard local, spherical averages. As an example, a local density dependent potential is fitted to reproduce the Hugoniot curve of a model of nitromethane over a large range of pressures and densities.

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.

Coupling microscopic and mesoscopic scales to simulate chemical equilibrium between a nanometric carbon cluster and detonation products fluid.

This paper presents a new method to obtain chemical equilibrium properties of detonation products mixtures including a solid carbon phase. In this work, the solid phase is modelled through a mesoparticle immersed in the fluid, such that the heterogeneous character of the mixture is explicitly taken into account. Inner properties of the clusters are taken from an equation of state obtained in a previous work, and interaction potential between the nanocluster and the fluid particles is derived from all-atoms simulations using the LCBOPII potential (Long range Carbon Bond Order Potential II). It appears that differences in chemical equilibrium results obtained with this method and the "composite ensemble method" (A. Hervouet et al., J. Phys. Chem. B, 2008, 112.), where fluid and solid phases are considered as non-interacting, are not significant, underlining the fact that considering the inhomogeneity of such system is crucial.

Microscopic calculations of Hugoniot curves of neat triaminotrinitrobenzene (TATB) and of its detonation products.

We compute the Hugoniot curves of both neat triaminotrinitrobenzene (TATB) and its detonation products mixture using atomistic simulation tools. To compute the Hugoniot states, we adapted our sampling constraints in average (SCA) method (Maillet et al. Appl. Math. Res. eXpress 2009, 2008, abn004) to Monte Carlo simulations. For neat TATB, we show that the potential proposed by Rai (Rai et al. J. Chem. Phys. 2008, 129, 194510) is not accurate enough to predict the Hugoniot curve and requires some optimization of its parameters. Concerning the detonation products, thermodynamic properties at chemical equilibrium are computed using a specific reaction ensemble Monte Carlo (RxMC) method (Bourasseau et al. Phys. Chem. Chem. Phys. 2011, 13, 7060), taking into account the presence of carbon clusters in the fluid mixture. We show that this explicit description of the solid phase immersed in the fluid phase modifies the chemical equilibrium.

Reinforcing materials modelling by encoding the structures of defects in crystalline solids into distortion scores.

This work revises the concept of defects in crystalline solids and proposes a universal strategy for their characterization at the atomic scale using outlier detection based on statistical distances. The proposed strategy provides a generic measure that describes the distortion score of local atomic environments. This score facilitates automatic defect localization and enables a stratified description of defects, which allows to distinguish the zones with different levels of distortion within the structure. This work proposes applications for advanced materials modelling ranging from the surrogate concept for the energy per atom to the relevant information selection for evaluation of energy barriers from the mean force. Moreover, this concept can serve for design of robust interatomic machine learning potentials and high-throughput analysis of their databases. The proposed definition of defects opens up many perspectives for materials design and characterization, promoting thereby the development of novel techniques in materials science.

Machine Learning Force Fields and Coarse-Grained Variables in Molecular Dynamics: Application to Materials and Biological Systems.

Machine learning encompasses tools and algorithms that are now becoming popular in almost all scientific and technological fields. This is true for molecular dynamics as well, where machine learning offers promises of extracting valuable information from the enormous amounts of data generated by simulation of complex systems. We provide here a review of our current understanding of goals, benefits, and limitations of machine learning techniques for computational studies on atomistic systems, focusing on the construction of empirical force fields from ab initio databases and the determination of reaction coordinates for free energy computation and enhanced sampling.

Ab initio simulations of thermodynamic and chemical properties of detonation product mixtures.

Thermodynamic and chemical properties of simple fluids N(2), CO(2), and H(2)O and their binary and ternary mixtures have been studied using density functional theory simulations in a high pressure and high temperature regime. We show that N(2) and binary mixtures with N(2) follow an ideal behavior over a large temperature and pressure range. On the contrary, the water molecule is observed to dissociate as either pressure or temperature increases. Dramatic consequences are observed when water is mixed with carbon dioxide at extreme conditions. Indeed, a new molecule is formed, CO(3)H(2), and the thermodynamic behavior of the mixture strongly deviates from ideality. Chemistry occurring at extreme conditions is then discussed in the context of detonation product modeling.

Constant entropy sampling and release waves of shock compressions.

We present or recall several equilibrium methods that allow one to compute isentropic processes, either during the compression or the release of the material. These methods are applied to compute the isentropic release of a shocked monoatomic liquid at high pressure and temperature. Moreover, equilibrium results of isentropic release are compared to the direct nonequilibrium simulation of the same process. We show that due to the viscosity of the liquid but also to nonequilibrium effects, the release of the system is not strictly isentropic.

Microscopic approaches to liquid nitromethane detonation properties.

In this paper, thermodynamic and chemical properties of nitromethane are investigated using microscopic simulations. The Hugoniot curve of the inert explosive is computed using Monte Carlo simulations with a modified version of the adaptative Erpenbeck equation of state and a recently developed intermolecular potential. Molecular dynamic simulations of nitromethane decomposition have been performed using a reactive potential, allowing the calculation of kinetic rate constants and activation energies. Finally, the Crussard curve of detonation products as well as thermodynamic properties at the Chapman-Jouguet (CJ) point are computed using reactive ensemble Monte Carlo simulations. Results are in good agreement with both thermochemical calculations and experimental measurements.

Thermodynamic behavior of the CO2 + NO2/N2O4 mixture: a Monte Carlo simulation study.

The thermodynamic behavior of the carbon dioxide + nitrogen dioxide (CO2 + NO2) mixture was investigated using a Monte Carlo molecular simulation approach. This system is a particularly challenging one because nitrogen dioxide exists as a mixture of monomers (NO2) and dimers (N2O4) under certain pressure and temperature conditions. The chemical equilibrium between N2O4 and 2NO2 and the vapor-liquid equilibrium of CO2 + NO2/N2O4 mixtures were simulated using simultaneously the reaction ensemble and the Gibbs ensemble Monte Carlo (RxMC and GEMC) methods. Rigid all atoms molecular potentials bearing point charges were proposed to model both NO2 and N2O4 species. Liquid-vapor coexistence properties of the reacting NO2/N2O4 system were first investigated. The calculated vapor pressures and coexisting densities were compared to experimental values, leading to an average deviation of 10% for vapor pressures and 6% for liquid densities. The critical region was also addressed successfully using the subcritical Monte Carlo simulation results and some appropriate scaling laws. Predictions of CO2 + NO2/N2O4 phase diagrams at 300, 313, and 330 K were then proposed. Derivative properties calculations were also performed in the reaction ensemble at constant pressure and temperature for both NO2/N2O4 and CO2 + NO2/N2O4 systems. The calculated heat capacities show a maximum in the temperature range where N2O4 dissociation occurs, in agreement with available experimental data.

Thermodynamic properties of benzene under shock conditions.

We present in this paper a thermodynamic analysis of benzene properties under shock conditions as given by molecular dynamics (MD) simulations. Reactive MD simulations of benzene predict a decomposition threshold corresponding to the flection point on the experimental Hugoniot curve. A polymerlike carbonated structure is observed for pressures above this threshold, but the calculated Hugoniot curve is in disagreement with the experimental one at high pressure. On the contrary, a system consisting of a diamond cluster in hydrogen gas leads to a correct prediction of the pressure on the Hugoniot curves. The central question is then linked to the kinetics of the transition between the polymerlike structure and the diamond cluster.

Size consistency in smoothed dissipative particle dynamics.

Smoothed dissipative particle dynamics (SDPD) is a mesoscopic method that allows one to select the level of resolution at which a fluid is simulated. In this work, we study the consistency of the resulting thermodynamic properties as a function of the size of the mesoparticles, both at equilibrium and out of equilibrium. We also propose a reformulation of the SDPD equations in terms of energy variables. This increases the similarities with dissipative particle dynamics with energy conservation and opens the way for a coupling between the two methods. Finally, we present a numerical scheme for SDPD that ensures the conservation of the invariants of the dynamics. Numerical simulations illustrate this approach.

Molecular based equation of state for shocked liquid nitromethane.

An approach is proposed to obtain the equation of state of unreactive shocked liquid nitromethane. Unlike previous major works, this equation of state is not based on extended integration schemes [P.C. Lysne, D.R. Hardesty, Fundamental equation of state of liquid nitromethane to 100 kbar, J. Chem. Phys. 59 (1973) 6512]. It does not follow the way proposed by Winey et al. [J.M. Winey, G.E. Duvall, M.D. Knudson, Y.M. Gupta, Equation of state and temperature measurements for shocked nitromethane, J. Chem. Phys. 113 (2000) 7492] where the specific heat C(v), the isothermal bulk modulus B(T) and the coefficient of thermal pressure (deltaP/deltaT)(v) are modeled as functions of temperature and volume using experimental data. In this work, we compute the complete equation of state by microscopic calculations. Indeed, by means of Monte Carlo molecular simulations, we have proposed a new force field for nitromethane that lead to a good description of shock properties [N. Desbiens, E. Bourasseau, J.-B. Maillet, Potential optimization for the calculation of shocked liquid nitromethane properties, Mol. Sim. 33 (2007) 1061; A. Hervouët, N. Desbiens, E. Bourasseau, J.-B. Maillet, Microscopic approaches to liquid nitromethane detonation properties, J. Phys. Chem. B 112 (2008) 5070]. Particularly, it has been shown that shock temperatures and second shock temperatures are accurately reproduced which is significative of the quality of the potential. Here, thermodynamic derivative properties are computed: specific heats, Grüneisen parameter, sound velocity among others, along the Hugoniot curve. This work constitutes to our knowledge the first determination of the equation of state of an unreactive shocked explosive by molecular simulations.

Molecular simulations of Hugoniots of detonation product mixtures at chemical equilibrium: microscopic calculation of the Chapman-Jouguet state.

In this work, we used simultaneously the reaction ensemble Monte Carlo (ReMC) method and the adaptive Erpenbeck equation of state (AE-EOS) method to directly calculate the thermodynamic and chemical equilibria of mixtures of detonation products on the Hugoniot curve. The ReMC method [W. R. Smith and B. Triska, J. Chem. Phys. 100, 3019 (1994)] allows us to reach the chemical equilibrium of a reacting mixture, and the AE-EOS method [J. J. Erpenbeck, Phys. Rev. A 46, 6406 (1992)] constrains the system to satisfy the Hugoniot relation. Once the Hugoniot curve of the detonation product mixture is established, the Chapman-Jouguet (CJ) state of the explosive can be determined. A NPT simulation at P(CJ) and T(CJ) is then performed in order to calculate direct thermodynamic properties and the following derivative properties of the system using a fluctuation method: calorific capacities, sound velocity, and Gruneisen coefficient. As the chemical composition fluctuates, and the number of particles is not necessarily constant in this ensemble, a fluctuation formula has been developed to take into account the fluctuations of mole number and composition. This type of calculation has been applied to several usual energetic materials: nitromethane, tetranitromethane, hexanitroethane, PETN, and RDX.

Atomistic mechanism for hot spot initiation.

We propose a picture of the role of shock-wave interactions with microscopic voids that leads to significant heating, sufficient to thermally initiate chemical reactions in solid explosives, or phase transitions in metals. The key ingredients to this dramatic overshoot in temperature are: (i) a strong enough shock wave to cause vaporization of material into the void; (ii) the stagnation of low-density vapor (for a wide enough gap) at the far side; and (iii) recompression of the gas (pressure-volume work) from low density back to the original shocked density. We explore dependencies on both shock strength and one-dimensional gap width in atomistic simulations of a two-dimensional unreactive Lennard-Jones solid, comparing observed thermal overshoot with a straightforward model, to show how hot spots can be generated under shock-wave conditions.