Ghanshyam Pilania

Accelerating materials property predictions using machine learning

The materials discovery process can be significantly expedited and simplified if we can learn effectively from available knowledge and data. In the present contribution, we show that efficient and accurate prediction of a diverse set of properties of material systems is possible by employing machine (or statistical) learning methods trained on quantum mechanical computations in combination with the notions of chemical similarity. Using a family of one-dimensional chain systems, we present a general formalism that allows us to discover decision rules that establish a mapping between easily accessible attributes of a system and its properties. It is shown that fingerprints based on either chemo-structural (compositional and configurational information) or the electronic charge density distribution can be used to make ultra-fast, yet accurate, property predictions. Harnessing such learning paradigms extends recent efforts to systematically explore and mine vast chemical spaces, and can significantly accelerate the discovery of new application-specific materials.

Machine learning bandgaps of double perovskites.

The ability to make rapid and accurate predictions on bandgaps of double perovskites is of much practical interest for a range of applications. While quantum mechanical computations for high-fidelity bandgaps are enormously computation-time intensive and thus impractical in high throughput studies, informatics-based statistical learning approaches can be a promising alternative. Here we demonstrate a systematic feature-engineering approach and a robust learning framework for efficient and accurate predictions of electronic bandgaps of double perovskites. After evaluating a set of more than 1.2 million features, we identify lowest occupied Kohn-Sham levels and elemental electronegativities of the constituent atomic species as the most crucial and relevant predictors. The developed models are validated and tested using the best practices of data science and further analyzed to rationalize their prediction performance.

Machine learning strategy for accelerated design of polymer dielectrics

The ability to efficiently design new and advanced dielectric polymers is hampered by the lack of sufficient, reliable data on wide polymer chemical spaces, and the difficulty of generating such data given time and computational/experimental constraints. Here, we address the issue of accelerating polymer dielectrics design by extracting learning models from data generated by accurate state-of-the-art first principles computations for polymers occupying an important part of the chemical subspace. The polymers are ‘fingerprinted’ as simple, easily attainable numerical representations, which are mapped to the properties of interest using a machine learning algorithm to develop an on-demand property prediction model. Further, a genetic algorithm is utilised to optimise polymer constituent blocks in an evolutionary manner, thus directly leading to the design of polymers with given target properties. While this philosophy of learning to make instant predictions and design is demonstrated here for the example of polymer dielectrics, it is equally applicable to other classes of materials as well.

Revisiting the Al/Al?O? interface: Coherent interfaces and misfit accommodation

We study the coherent and semi-coherent Al/α-Al2O3 interfaces using molecular dynamics simulations with a mixed, metallic-ionic atomistic model. For the coherent interfaces, both Al-terminated and O-terminated nonstoichiometric interfaces have been studied and their relative stability has been established. To understand the misfit accommodation at the semi-coherent interface, a 1-dimensional (1D) misfit dislocation model and a 2-dimensional (2D) dislocation network model have been studied. For the latter case, our analysis reveals an interface dislocation structure with a network of three sets of parallel dislocations, each with pure-edge character, giving rise to a pattern of coherent and stacking-fault-like regions at the interface. Structural relaxation at elevated temperatures leads to a further change of the dislocation pattern, which can be understood in terms of a competition between the stacking fault energy and the dislocation interaction energy at the interface. Our results are expected to serve as an input for the subsequent dislocation dynamics models to understand and predict the macroscopic mechanical behavior of Al/α-Al2O3 composite heterostructures.

New Group IV Chemical Motifs for Improved Dielectric Permittivity of Polyethylene

An enhanced dielectric permittivity of polyethylene and related polymers, while not overly sacrificing their excellent insulating properties, is highly desirable for various electrical energy storage applications. In this computational study, we use density functional theory (DFT) in combination with modified group additivity based high throughput techniques to identify promising chemical motifs that can increase the dielectric permittivity of polyethylene. We consider isolated polyethylene chains and allow the CH2 units in the backbone to be replaced by a number of Group IV halides (viz., SiF2, SiCl2, GeF2, GeCl2, SnF2, or SnCl2 units) in a systematic, progressive, and exhaustive manner. The dielectric permittivity of the chemically modified polyethylene chains is determined by employing DFT computations in combination with the effective medium theory for a limited set of compositions and configurations. The underlying chemical trends in the DFT data are first rationalized in terms of various tabulated atomic properties of the constituent atoms. Next, by parametrizing a modified group contribution expansion using the DFT data set, we are able to predict the dielectric permittivity and bandgap of nearly 30,000 systems spanning a much larger part of the configurational and compositional space. Promising motifs which lead to simultaneously large dielectric constant and band gap in the modified polyethylene chains have been identified. Our theoretical work is expected to serve as a possible motivation for future experimental efforts.

A polymer dataset for accelerated property prediction and design

Emerging computation- and data-driven approaches are particularly useful for rationally designing materials with targeted properties. Generally, these approaches rely on identifying structure-property relationships by learning from a dataset of sufficiently large number of relevant materials. The learned information can then be used to predict the properties of materials not already in the dataset, thus accelerating the materials design. Herein, we develop a dataset of 1,073 polymers and related materials and make it available at http://khazana.uconn.edu/. This dataset is uniformly prepared using first-principles calculations with structures obtained either from other sources or by using structure search methods. Because the immediate target of this work is to assist the design of high dielectric constant polymers, it is initially designed to include the optimized structures, atomization energies, band gaps, and dielectric constants. It will be progressively expanded by accumulating new materials and including additional properties calculated for the optimized structures provided.

Termination chemistry-driven dislocation structure at SrTiO3/MgO heterointerfaces

Exploiting the promise of nanocomposite oxides necessitates a detailed understanding of the dislocation structure at the interfaces, which governs diverse and technologically relevant properties. Here we report atomistic simulations demonstrating a strong dependence of the dislocation structure on the termination chemistry at the SrTiO3/MgO heterointerface. The SrO- and TiO2-terminated interfaces exhibit distinct nearest neighbour arrangements between cations and anions, leading to variations in local electrostatic interactions across the interface that ultimately dictate the dislocation structure. Networks of dislocations with different Burgers vectors and dislocation spacing characterize the two interfaces. These networks in turn influence the overall stability of and the behaviour of oxygen vacancies at the heterointerface, which will dictate vital properties such as mass transport at the interface. To date, the observed correlation between the dislocation structure and the termination chemistry at the interface has not been recognized, and offers novel avenues for fine-tuning oxide nanocomposites with enhanced functionalities.

Machine learning in materials informatics: recent applications and prospects

Propelled partly by the Materials Genome Initiative, and partly by the algorithmic developments and the resounding successes of data-driven efforts in other domains, informatics strategies are beginning to take shape within materials science. These approaches lead to surrogate machine learning models that enable rapid predictions based purely on past data rather than by direct experimentation or by computations/simulations in which fundamental equations are explicitly solved. Data-centric informatics methods are becoming useful to determine material properties that are hard to measure or compute using traditional methods—due to the cost, time or effort involved—but for which reliable data either already exists or can be generated for at least a subset of the critical cases. Predictions are typically interpolative, involving fingerprinting a material numerically first, and then following a mapping (established via a learning algorithm) between the fingerprint and the property of interest. Fingerprints, also referred to as “descriptors”, may be of many types and scales, as dictated by the application domain and needs. Predictions may also be extrapolative—extending into new materials spaces—provided prediction uncertainties are properly taken into account. This article attempts to provide an overview of some of the recent successful data-driven “materials informatics” strategies undertaken in the last decade, with particular emphasis on the fingerprint or descriptor choices. The review also identifies some challenges the community is facing and those that should be overcome in the near future.

Finding new perovskite halides via machine learning

Advanced materials with improved properties have the potential to fuel future technological advancements. However, identification and discovery of these optimal materials for a specific application is a non-trivial task, because of the vastness of the chemical search space with enormous compositional and configurational degrees of freedom. Materials informatics provides an efficient approach towards rational design of new materials, via learning from known data to make decisions on new and previously unexplored compounds in an accelerated manner. Here, we demonstrate the power and utility of such statistical learning (or machine learning) via building a support vector machine (SVM) based classifier that uses elemental features (or descriptors) to predict the formability of a given ABX3 halide composition (where A and B represent monovalent and divalent cations, respectively, and X is F, Cl, Br or I anion) in the perovskite crystal structure. The classification model is built by learning from a dataset of 181 experimentally known ABX3 compounds. After exploring a wide range of features, we identify ionic radii, tolerance factor and octahedral factor to be the most important factors for the classification, suggesting that steric and geometric packing effects govern the stability of these halides. The trained and validated models then predict, with a high degree of confidence, several novel ABX3 compositions with perovskite crystal structure.

Classification of ABO3 perovskite solids: a machine learning study.

We explored the use of machine learning methods for classifying whether a particular ABO3 chemistry forms a perovskite or non-perovskite structured solid. Starting with three sets of feature pairs (the tolerance and octahedral factors, the A and B ionic radii relative to the radius of O, and the bond valence distances between the A and B ions from the O atoms), we used machine learning to create a hyper-dimensional partial dependency structure plot using all three feature pairs or any two of them. Doing so increased the accuracy of our predictions by 2-3 percentage points over using any one pair. We also included the Mendeleev numbers of the A and B atoms to this set of feature pairs. Doing this and using the capabilities of our machine learning algorithm, the gradient tree boosting classifier, enabled us to generate a new type of structure plot that has the simplicity of one based on using just the Mendeleev numbers, but with the added advantages of having a higher accuracy and providing a measure of likelihood of the predicted structure.