Search

Motivated by the problem of optimization of force-field systems in physics using large-scale computer simulations, we consider exploration of a deterministic complex multivariate response surface. The objective is to find input combinations that generate output close to some desired or "target" vector. In spite of reducing the problem to exploration of the input space with respect to a one-dimensional loss function, the search is nontrivial and challenging due to infeasible input combinations, high dimensionalities of the input and output space and multiple "desirable" regions in the input space and the difficulty of emulating the objective function well with a surrogate model. We propose an approach that is based on combining machine learning techniques with smart experimental design ideas to locate multiple good regions in the input space.

The surface dissolution at the LATP aqueous interface is a sequentially dynamic process starting with phosphate dissolution destabilizing the NASICON structure and continues with a titanium-rich secondary phase formation.

International audience; This study reports the sintering of zinc oxide (ZnO) through the comparison between the hydrothermal sintering (HS) and the cold sintering process (CSP) operating in closed and open conditions, respectively. Sintering was performed at 155 ± 5 °C applying a pressure of 320 MPa, and during different holding times (0 min, 20 min, 40 min and 80 min). Whatever the low sintering process used, ceramics characteristics are almost similar in terms of relative densities and ZnO structure. However, several differences such as the nature of stabilized phases, grain sizes and quantities of residual molecules in the densified pellets, were characterized and explained. The formation of zinc acetate “bridges” was observed ex situ in hydrothermally sintered samples. A detailed ReaxFF molecular dynamics simulation was performed to help understand the formation mechanisms of zinc acetate “bridges” and compare the chemical activities between HS and CSP.

Molecular dynamics (MD) simulations play a crucial role in modeling biomolecular systems in which the electrostatic interactions are critical in dictating the structural and dynamical properties. Thus, the treatment of the electrostatic interactions defined in the underlying force field (FF) strongly affects the simulation accuracy. Most FFs use fixed partial atomic charges to include electrostatic interactions, and therefore lack the electronic polarization response, representing an intrinsic limitation. To address this limitation, polarizable FFs have been developed that treat atomic polarizabilities explicitly. Here we present the application of the all-atom polarizable (Drude) and non-polarizable (CHARMM) nucleic acid FFs in RNA hairpin systems to investigate the impact of polarization on structural properties, dipole moment distributions, and cation interactions. Results show that the presence of polarizability in the FF significantly improves the stabilization of RNA hairpin structure. As expected, the distributions of dipole moments show more fluctuations when simulated using the polarizable FF, with the variation in dipoles contributing to the stabilization of the structures of the loop regions of the RNAs. Contact map analyses between the bases and cations show that the variation of the ion distribution around the entire hairpin is larger for the polarizable FF and the cations occupy the outer hydration shell to a greater extent. The presented results indicate the importance of the explicit treatment of electronic polarizability in molecular simulations of RNA, including in non-canonical regions.

The cold sintering process (CSP) densifies ceramics at much lower temperatures than conventional sintering processes. Several ceramics and composite systems have been successfully densified under cold sintering. For the grain growth kinetics of zinc oxide, reduced activation energies are shown, and yet the mechanism behind this growth is unknown. Herein, we investigate these mechanisms in more detail with experiments and ReaxFF molecular dynamics simulations. We investigated the recrystallization of zinc cations under various acidic conditions and found that their adsorption to the surface can be a rate-limiting factor for cold sintering. Our studies show that surface hydroxylation in CSP does not inhibit crystallization; in contrast, by creating a surface complex, it creates an orders of magnitude acceleration in surface diffusion, and in turn, accelerates recrystallization.

Molten hydroxides, often used for crystal growth and nanoparticle synthesis, have recently been applied for the single step densification of several inorganic materials under moderate uniaxial pressures and 1000 °C below their usual sintering temperatures. The latter approach, termed cold sintering process (CSP), is a mechanochemically driven process that enables the densification of inorganic materials through a dissolution-precipitation creep mechanism. In this study, we report the main densification mechanisms of BaTiO3 in a NaOH-KOH eutectic mixture. A chemical insight at the atomistic level, investigated by ReaxFF molecular dynamics simulations, offers plausible ionic complex formation scenarios and reactions at the BaTiO3/molten hydroxide interface, enabling the dissolution-precipitation reactions and the subsequent cold sintering of BaTiO3.

ReaxFF is an empirical interatomic potential capable of simulating reactions in complex chemical processes and thus determine the dynamical evolution of the molecular systems. A drawback of this method is the necessity of a significant preprocessing to adapt it to a chemical system of nterest. One of the preprocessing steps is the optimization of force field parameters that are used to tune interatomic interactions to mimic ones obtained by quantum chemistry-based methods. The optimization of these parameters is a very complex high dimensional problem. Here, we propose an INitial-DEsign Enhanced Deep learning-based OPTimization (INDEEDopt) framework to be used in ReaxFF parameterization. The procedure starts with a Latin Hypercube Design (LHD) algorithm that is used to explore the parameter landscape extensively. The LHD passes the information about explored regions to a deep learning model for training. The deep learning model finds the minimum discrepancy regions and eliminates unfeasible regions, which originate from the unphysical atomistic interactions, and constructs a more comprehensive understanding of a physically meaningful parameter space. The procedure was successfully used to parametrize a nickel-chromium binary force field and a tungsten-sulfide-carbon-hydrogen quaternary force field and produced improved accuracies in shorter periods time compared to conventional optimization method.

We developed a ReaxFF reactive force field for NASICON-type Li1+xAlxTi2−x(PO4)3 (LATP) materials, which is a promising solid-electrolyte that may enable all-solid-state lithium-ion batteries. The force field parameters were optimized based on density functional theory (DFT) data, including equations of state and the heats of formation of ternary metal oxides and metal phosphate crystal phases (e.g., LixTiO2, Al2TiO5, LiAlO2, AlPO4, Li3PO4 and LiTi2(PO4)3 (LTP)), and the energy barriers for Li diffusion in TiO2 and LTP via vacancies and interstitial sites. Using ReaxFF, the structural and the energetic features of LATP were described properly across various compositions – Li occupies more preferentially the interstitial site next to Al than next to Ti. Also, as observed in experimental data, the lattice parameters decrease when Ti is partly substituted by Al because of the smaller size of the Al cation. Using this force field, the diffusion mechanism and the ionic conductivity of Li in LTP and LATP were investigated at T = 300–1100 K. Low ionic conductivity (5.9 × 10−5 S cm−1 at 300 K) was obtained in LTP as previously reported. In LATP at x = 0.2, the ionic conductivity was slightly improved (8.4 × 10−5 S cm−1), but it is still below the experimental value, which is on the order of 10−4 to 10−3 S cm−1 at x = 0.3–0.5. At higher x (higher Al composition), LATP has a configurational diversity due to the Al substitution and the concomitant insertion of Li. By performing a hybrid MC/MD simulation for LATP at x = 0.5, a thermodynamically stable LATP configuration was obtained. The ionic conductivity of this LATP configuration was calculated to be 7.4 × 10−4 S cm−1 at 300 K, which is one order of magnitude higher than the ionic conductivity for LTP and LATP at x = 0.2. This value is in good agreement with our experimental value (2.5 × 10−4 S cm−1 at 300 K) and the literature values. The composition-dependent ionic conductivity of LATP was successfully demonstrated using the ReaxFF reactive force field, verifying the applicability of the LATP force field for the understanding of Li diffusion and the design of highly Li ion conductive solid electrolytes. Furthermore, our results also demonstrate the feasibility of the MC/MD method in modeling LATP configuration, and provide compelling evidence for the solid solution sensitivity on ionic conductivity.

Duckweed has emerged as a potential feedstock for the environmentally sustainable and economically viable production of biofuels and protein. This aim of this study was to: (1) enhance an existing intrinsic duckweed growth model and apply it for developing a general regression model that enables users to easily predict annual duckweed yield for large scale applications; and (2) determine the optimal parameter sets that produce the highest annual duckweed yield at a specific location for known values of daily temperature and photoperiod. Simulations performed using Stella Architect were used to compute annual duckweed yield and generate separate datasets for developing a regression model and optimization model. To improve duckweed yields for large-scale applications which incorporate regular harvesting, a harvesting regime was added to the intrinsic duckweed growth model. Two new parameters (harvest frequency and harvest ratio) and a control (harvest threshold) were used to describe the harvesting regime in the model. The general model was developed by fitting LASSO regression (R2 = 0.95) with four variables: initial mat density, intrinsic growth rate, harvest ratio, and harvest frequency. This model offers a simple method for users to estimate annual duckweed yield in practical applications without the need for dynamic simulation runs. Optimum parameter values to maximize biomass production at a location in southwest Florida, USA, were determined using an optimization framework involving a deep neural network machine learning algorithm. Using an existing daylength model to predict daily photoperiod and inputting local temperature data, machine learning calculated a maximum yield of 70 dry tons per hectare per year for the Florida case study, under the following conditions: initial mat density = 169 gdry m−2; harvest threshold = 76 gdry m−2; nitrogen = 50.1 mg L−1; phosphorus = 7.5 mg L−1; harvest ratio = 0.35; and harvest frequency = 1 day.

The cold sintering process (CSP) is a technique that enables densification of ceramics and composites at extremely low temperatures (T ​

The reaction dynamics of a liquid-solid interface with the example of an acetic acid/water solution interacting with a ZnO(1010) surface was investigated using ReaxFF reactive force field-based molecular dynamics. The interactions were studied over a broad temperature range to assess the kinetics and reaction pathways. Two different acetic acid dissociation mechanisms are observed in the simulations: (1) deprotonation to surface cation, which produces a terminal hydroxyl and (2) deprotonation to a bridging hydroxyl, which results in water production. An increase in temperature promotes the dissociation of acetic acids and its adsorption to surface at first, but as the temperature increase continues, the surface coverage by acetates decreases due to evaporation from the surface or decomposition. The acetate decomposition starts with a nucleophilic attack of oxygen to methyl carbon and results in the production of carbon dioxide, which is consistent with experimental findings in the literature. The coverage of the surface by water molecules decreases as the system is heated up, which is also observed in other molecular dynamics studies. At elevated temperatures, acetate molecules are more stable than water molecules or bridging hydroxyls on the surface. These simulations validate the ReaxFF method for the water/organic mixture and metal oxide surface interactions and provide insights into structure and reactivity of aqueous solvents on metal oxide surfaces at elevated temperatures. Adsorption trends that are observed in this study are consistent with phenomenological Langmuir models. The reaction of acetic acid decomposition to smaller molecules such as CO2 and CH2O agrees with experimental observations. Understanding the details of these dynamic surface reactions are critical to understand important new cold sintering processes that utilize transient liquid and solid reactions, and the latter could be used to predict solvent selection for cold sintering.

We developed a ReaxFF reactive force field for NASICON-type Li1+xAlxTi2-x(PO4)3 (LATP) materials, which is a promising solid-electrolyte that may enable all-solid-state lithium-ion batteries. The force field parameters were optimized based on density functional theory (DFT) data, including equations of state and the heats of formation of ternary metal oxides and metal phosphate crystal phases (e.g., LixTiO2, Al2TiO5, LiAlO2, AlPO4, Li3PO4 and LiTi2(PO4)3 (LTP)), and the energy barriers for Li diffusion in TiO2 and LTP via vacancies and interstitial sites. Using ReaxFF, the structural and the energetic features of LATP were described properly across various compositions - Li occupies more preferentially the interstitial site next to Al than next to Ti. Also, as observed in experimental data, the lattice parameters decrease when Ti is partly substituted by Al because of the smaller size of the Al cation. Using this force field, the diffusion mechanism and the ionic conductivity of Li in LTP and LATP were investigated at T = 300-1100 K. Low ionic conductivity (5.9 × 10-5 S cm-1 at 300 K) was obtained in LTP as previously reported. In LATP at x = 0.2, the ionic conductivity was slightly improved (8.4 × 10-5 S cm-1), but it is still below the experimental value, which is on the order of 10-4 to 10-3 S cm-1 at x = 0.3-0.5. At higher x (higher Al composition), LATP has a configurational diversity due to the Al substitution and the concomitant insertion of Li. By performing a hybrid MC/MD simulation for LATP at x = 0.5, a thermodynamically stable LATP configuration was obtained. The ionic conductivity of this LATP configuration was calculated to be 7.4 × 10-4 S cm-1 at 300 K, which is one order of magnitude higher than the ionic conductivity for LTP and LATP at x = 0.2. This value is in good agreement with our experimental value (2.5 × 10-4 S cm-1 at 300 K) and the literature values. The composition-dependent ionic conductivity of LATP was successfully demonstrated using the ReaxFF reactive force field, verifying the applicability of the LATP force field for the understanding of Li diffusion and the design of highly Li ion conductive solid electrolytes. Furthermore, our results also demonstrate the feasibility of the MC/MD method in modeling LATP configuration, and provide compelling evidence for the solid solution sensitivity on ionic conductivity.

The intermolecular structure formation in liquid and supercritical acetic acid-water mixtures was investigated using ReaxFF-based molecular dynamics simulations. The microscopic structures of acetic acid-water mixtures with different acetic acid mole fractions (1.0 ≥ xHAc ≥ 0.2) at ambient and critical conditions were examined. The potential energy surface associated with the dissociation of acetic acid molecules was calculated using a metadynamics procedure to optimize the dissociation energy of ReaxFF potential. At ambient conditions, depending on the acetic acid concentration, either acetic acid clusters or water clusters are dominant in the liquid mixture. When acetic acid is dominant (0.4 ≤ xHAc), cyclic dimers and chain structures between acetic acid molecules are present in the mixture. Both structures disappear at increased water content of the mixture. It was found by simulations that the acetic acid molecules released from these dimer and chain structures tend to stay in a dipole-dipole interaction. These structural changes are in agreement with the experimental results. When switched to critical conditions, the long-range interactions (e.g., second or fourth neighbor) disappear and the water-water and acetic acid-acetic acid structural formations become disordered. The simulated radial distribution function for water-water interactions is in agreement with experimental and computational studies. The first neighbor interactions between acetic acid and water molecules are preserved at relatively lower temperatures of the critical region. As higher temperatures are reached in the critical region, these interactions were observed to weaken. These simulations indicate that ReaxFF molecular dynamics simulations are an appropriate tool for studying supercritical water/organic acid mixtures.