Tarak Patra

Polymer nanocomposites (PNCs) offer a broad range of thermophysical properties that are linked to their compositions. However, it is challenging to establish a universal composition-property relationship in PNCs due to their wide-ranging composition and chemical space. Here, we address this problem and develop a new method to model the composition-microstructure relation of a PNC through an intelligent machine-learning pipeline named nanoNET. The nanoNET is a nanoparticles (NPs) distribution predictor, built upon computer vision and image recognition concepts. It integrates unsupervised deep learning and regression in a fully automated pipeline. We conduct coarse-grained molecular dynamics simulations of PNCs and utilize the data to establish and validate the nanoNET. Within this framework, a random forest regression model predicts the distribution of NPs in a PNC in a latent space. Subsequently, a convolutional neural network-based decoder converts the latent space representation to the actual radial distribution function (RDF) of NPs in the given PNC. The nanoNET predicts NPs distribution in many unknown PNCs very accurately. This method is very generic and can accelerate the design, discovery, and fundamental understanding of composition-microstructure relationships in PNCs and other molecular systems.

While molecular simulations have contributed to the modern understanding of the glass transition, they are constrained in prediction of experimental glass temperatures Tg because they are limited to times far shorter than those associated with experimental glass formation. Here, we bridge this gap via a model-based forecasting approach. We assess models of the temperature dependence of dynamics in glass forming liquids based upon the rate at which their prediction of Tg and fragility converge upon incorporating increasingly long timescale data. We find that the Cooperative Model of Schmidtke et al. typically provides the best predictions, ultimately enabling good Tg predictions from all-atom simulations of a range of polymers. Together with a protocol for efficient simulation of dynamics in glass-forming liquids, this strategy enables high-throughput computational screening of the glass transition. The success of the Cooperative Model in predicting low temperature behavior adds support to the two-barrier scenario underlying this model.

Machine learning (ML) and artificial intelligence (AI) have remarkable abilities to classify, recognize, and characterize complex patterns and trends in large data sets. Here, we adopt a subclass of ML methods, viz., deep learning, and develop a general purpose AI tool - dPOLY - for analyzing molecular dynamics (MD) trajectories and predicting phases and phase transitions in polymers. An unsupervised deep neural network (DNN) is used within this framework to map a MD trajectory undergoing thermophysical treatment such as cooling, heating, drying, and compression to a lower dimension. A supervised DNN is subsequently developed based on the lower dimensional data to characterize the phases and phase transitions. As a proof of concept, we employ this framework to study the coil to globule phase transition of a model polymer system. We conduct coarse-grained MD simulations to collect MD trajectories of a single polymer chain over a wide range of temperatures and use the dPOLY framework to predict polymer phases. The dPOLY framework accurately predicts the critical temperatures for the coil to globule transition for a wide range of polymer sizes. This method is generic and can be extended to capture various other phase transitions and dynamical crossovers in polymers and other soft materials. It can also significantly accelerate polymer phase prediction and characterization.

The correlations between the sequence of monomers in a macromolecule and its three-dimensional (3D) structure is a grand challenge in polymer science. The properties and functions of macromolecules depend on their 3D shape that has appeared to be dictated by their monomer sequence. However, the progress towards understanding the sequence–structure-property correlations and their utilization in materials engineering are slow because it is almost impossible to characterize an astronomically large number of possible sequences of a copolymer using traditional experimental and simulation methods. To address this problem, here, we combine evolutionary computing and coarse-grained molecular dynamics (CGMD) simulation and study the sequence-structure correlations of a model AB-type copolymer in a solution and assess the impact of sequence on the packing density in its bulk phase. The CGMD-based evolutionary algorithm (EA) screens the sequence space of a single chain copolymer efficiently and identifies a wide range of single-molecule structures including extremal radii of gyration. The data are utilized to estimate the Pareto front of the structure-space of a binary copolymer as a function of its composition. The monomer packings in single-molecule solution phase and multimolecular bulk phase are found to be identical. The work highlights the opportunities of sequence-specific control of macromolecular structure for designing target materials.

Small metal clusters are of fundamental scientific interest and of tremendous significance in catalysis. These nanoscale clusters display diverse geometries and structural motifs depending on the cluster size; a knowledge of this size-dependent structural motifs and their dynamical evolution has been of longstanding interest. Given the high computational cost of first-principles calculations, molecular modeling and atomistic simulations such as molecular dynamics (MD) has proven to be an important complementary tool to aid this understanding. Classical MD typically employ predefined functional forms which limits their ability to capture such complex size-dependent structural and dynamical transformation. Neural Network (NN) based potentials represent flexible alternatives and in principle, well-trained NN potentials can provide high level of flexibility, transferability and accuracy on-par with the reference model used for training. A major challenge, however, is that NN models are interpolative and requires large quantities ((Formula presented.) or greater) of training data to ensure that the model adequately samples the energy landscape both near and far-from-equilibrium. A highly desirable goal is minimize the number of training data, especially if the underlying reference model is first-principles based and hence expensive. Here, we introduce an active learning (AL) scheme that trains a NN model on-the-fly with minimal amount of first-principles based training data. Our AL workflow is initiated with a sparse training dataset ((Formula presented.) 1 to 5 data points) and is updated on-the-fly via a Nested Ensemble Monte Carlo scheme that iteratively queries the energy landscape in regions of failure and updates the training pool to improve the network performance. Using a representative system of gold clusters, we demonstrate that our AL workflow can train a NN with (Formula presented.) 500 total reference calculations. Using an extensive DFT test set of ∼1100 configurations, we show that our AL-NN is able to accurately predict both the DFT energies and the forces for clusters of a myriad of different sizes. Our NN predictions are within 30 meV/atom and 40 meV/Å of the reference DFT calculations. Moreover, our AL-NN model also adequately captures the various size-dependent structural and dynamical properties of gold clusters in excellent agreement with DFT calculations and available experiments. We finally show that our AL-NN model also captures bulk properties reasonably well, even though they were not included in the training data.

Reinforcement learning (RL) algorithms that include Monte Carlo Tree Search (MCTS) have found tremendous success in computer games such as Go, Shiga and Chess. Such learning algorithms have demonstrated super-human capabilities in navigating through an exhaustive discrete action search space. Motivated by their success in computer games, we demonstrate that RL can be applied to inverse materials design problems. We deploy RL for a representative case of the optimal atomic scale inverse design of extended defects via rearrangement of chalcogen (e.g. S) vacancies in 2D transition metal dichalcogenides (e.g. MoS2 ). These defect rearrangements and their dynamics are important from the perspective of tunable phase transition in 2D materials i.e. 2H (semi-conducting) to 1T (metallic) in MoS2. We demonstrate the ability of MCTS interfaced with a reactive molecular dynamics simulator to efficiently sample the defect phase space and perform inverse design—starting from randomly distributed S vacancies, the optimal defect rearrangement of defects corresponds a line defect of S vacancies. We compare MCTS performance with evolutionary optimization i.e. genetic algorithms and show that MCTS converges to a better optimal solution (lower objective) and in fewer evaluations compared to GA. We also comprehensively evaluate and discuss the effect of MCTS hyperparameters on the convergence to solution. Overall, our study demonstrates the effectives of using RL approaches that operate in discrete action space for inverse defect design problems.

Lubricity, a phenomenon which enables the ease of motion of objects, and wear resistance, which minimizes material damage or degradation, are important fundamental characteristics for sustainable technology developments. Ultrathin coatings that promote lubricity and wear resistance are of huge importance for a number of applications, including magnetic storage and micro-/nanoelectromechanical systems. Conventional ultrathin coatings have, however, reached their limit. Graphene-based materials that have shown promise to reduce friction and wear have many intrinsic limitations such as high temperature and substrate-specific growth. To address these concerns, a great deal of research is currently ongoing to optimize graphene-based materials. Here we discover that angstrom-thick carbon (8 Å) significantly reduces interfacial friction and wear. This lubricant shows ultrahigh optical transparency and can be directly deposited on a wide range of surfaces at room temperature. Experiments combined with molecular dynamics simulations reveal that the lubricating efficacy of 8 Å carbon is further improved via interfacial nitrogen.

Grafting polymer chains on the surfaces of nanoparticles is a well-known route to control their self-assembly and distribution in a polymer matrix. A wide variety of self-assembled structures are achieved by changing the grafting patterns on the surface of an individual nanoparticle. However, an accurate estimation of the effective potential of mean force between a pair of grafted nanoparticles that determines their assembly and distribution in a polymer matrix is an outstanding challenge in nanoscience. We address this problem via deep learning. As a proof of concept, here we report a deep learning framework that learns the interaction between a pair of single-chain grafted spherical nanoparticles from their molecular dynamics trajectory. Subsequently, we carry out the deep learning potential of mean force-based molecular simulation that predicts the self-assembly of a large number of single-chain grafted nanoparticles into various anisotropic superstructures, including percolating networks and bilayers depending on the nanoparticle concentration in three-dimensions. The deep learning potential of mean force-predicted self-assembled superstructures are consistent with the actual superstructures of single-chain polymer grafted spherical nanoparticles. This deep learning framework is very generic and extensible to more complex systems including multiple-chain grafted nanoparticles. We expect that this deep learning approach will accelerate the characterization and prediction of the self-assembly and phase behaviour of polymer-grafted and unfunctionalized nanoparticles in free space or a polymer matrix.

Neural network (NN) based potentials represent flexible alternatives to pre-defined functional forms. Well-trained NN potentials are transferable and provide a high level of accuracy on-par with the reference model used for training. Despite their tremendous potential and interest in them, there are at least two challenges that need to be addressed-(1) NN models are interpolative, and hence trained by generating large quantities (~104 or greater) of structural data in hopes that the model has adequately sampled the energy landscape both near and far-from-equilibrium. It is desirable to minimize the number of training data, especially if the underlying reference model is expensive. (2) NN atomistic potentials (like any other classical atomistic model) are limited in the time scales they can access. Coarse-grained NN potentials have emerged as a viable alternative. Here, we address these challenges by introducing an active learning scheme that trains a CG model with a minimal amount of training data. Our active learning workflow starts with a sparse training data set (~1 to 5 data points), which is continually updated via a nested ensemble Monte Carlo scheme that iteratively queries the energy landscape in regions of failure and improves the network performance. We demonstrate that with ~300 reference data, our AL-NN is able to accurately predict both the energies and the molecular forces of water, within 2?meV per molecule and 40 meV Å-1 of the reference (coarse-grained bond-order potential) model. The AL-NN water model provides good prediction of several structural, thermodynamic, and temperature dependent properties of liquid water, with values close to those obtained from the reference model. The AL-NN also captures the well-known density anomaly of liquid water observed in experiments. Although the AL procedure has been demonstrated for training CG models with sparse reference data, it can be easily extended to develop atomistic NN models against a minimal amount of high-fidelity first-principles data.

The global plastics problem is a trifecta, greatly affecting environment, energy and climate1–4. Many innovative closed/open-loop plastics recycling or upcycling strategies have been proposed or developed5–16, addressing various aspects of the issues underpinning the achievement of a circular economy17–19. In this context, reusing mixed-plastics waste presents a particular challenge with no current effective closed-loop solution20. This is because such mixed plastics, especially polar/apolar polymer mixtures, are typically incompatible and phase separate, leading to materials with substantially inferior properties. To address this key barrier, here we introduce a new compatibilization strategy that installs dynamic crosslinkers into several classes of binary, ternary and postconsumer immiscible polymer mixtures in situ. Our combined experimental and modelling studies show that specifically designed classes of dynamic crosslinker can reactivate mixed-plastics chains, represented here by apolar polyolefins and polar polyesters, by compatibilizing them via dynamic formation of graft multiblock copolymers. The resulting in-situ-generated dynamic thermosets exhibit intrinsic reprocessability and enhanced tensile strength and creep resistance relative to virgin plastics. This approach avoids the need for de/reconstruction and thus potentially provides an alternative, facile route towards the recovery of the endowed energy and materials value of individual plastics.