MXenes are two-dimensional (2D) transition metal carbides and nitrides, and are invariably metallic in pristine form. While spontaneous passivation of their reactive bare surfaces lends unprecedented functionalities, consequently a many-folds increase in number of possible functionalized MXene makes their characterization difficult.Here, we study the electronic properties of this vast class of materials by accurately estimating the band gaps using statistical learning. Using easily available properties of the MXene, namely, boiling and melting points, atomic radii, phases, bond lengths, etc., as input features, models were developed using kernel ridge (KRR), support vector (SVR), Gaussian process (GPR), and bootstrap aggregating regression algorithms. Among these, the GPR model predicts the band gap with lowest root-mean-squared error (rmse) of 0.14 eV, within seconds. Most importantly, these models do not involve the Perdew−Burke−Ernzerhof (PBE) band gap as a feature. Our results demonstrate that machine-learning models can bypass the band gap underestimation problem of local and semilocal functionals used in density functional theory (DFT) calculations, without subsequent correction using the time-consuming GW approach.

Low thermal conductivity materials are crucial for applications such as thermoelectric conversion of waste heat to useful energy and thermal barrier coatings. On the other hand, high thermal conductivity materials are necessary for cooling electronic devices. However, search for such materials via explicit evaluation of thermal conductivity either experimentally or computationally is very challenging. Here, we carried out high-throughput ab initio calculations, on a dataset containing 195 binary, ternary, and quaternary compounds. The lattice thermal conductivity κ l values of 120 dynamically stable and nonmetallic compounds are calculated, which span over 3 orders of magnitude. Among these, 11 ultrahigh and 15 ultralow κ l materials are identified. An analysis of generated property map of this dataset reveals a strong dependence of κ l on simple descriptors, namely, maximum phonon frequency, integrated Gruneisen parameter up to 3 THz, average atomic mass, and volume of the unit cell. Using these descriptors, a Gaussian process regression-based machine learning (ML) model is developed. The model predicts log-scaled κ l with a very small root mean square error of ∼0.21. Comparatively, the Slack model, which uses more involved parameters, severely overestimates κ l . The superior performance of our ML model can ensure a reliable and accelerated search for multitude of low and high thermal conductivity materials.

Functionalized MXene has emerged a promising class of two-dimensional materials having more than tens of thousands of compounds, whose uses may range from electronics to energy applications. Other than the band gap, these properties rely on the accurate position of the band edges. Hence, to synthesize MXenes for various applications, a prior knowledge of the accurate position of their band edges at an absolute scale is essential; computing these with conventional methods would take years for all the MXenes. Here, we develop a machine learning model for positioning the band edges with GW level of accuracy having a minimum root-mean-squared error of 0.12 eV. An intuitive model is proposed based on the combination of Perdew–Burke–Ernzerhof band edge and vacuum potential having a correlation of 0.93 with GW band edges. These models can be utilized to identify MXenes for a desired application in an accelerated manner.

Efficiency of a thermoelectric material relies on a combination of electronic and thermal transport properties, which are governed by various scattering mechanisms. Explicit evaluation of temperature dependent scattering time or the electron relaxation time (τ el ) is thus necessary to assess the efficiency of thermoelectrics. Experimental or computational measurement of τ el is very challenging due to the inherent time limitation and high computational cost. Herein, a statistical machine learning (ML) based approach has been developed to predict the experimental electrical conductivity (σ) followed by an estimation of the relaxation time (τ el ). By utilizing a unique mean ranking method for feature selection, simple elemental properties such as the boiling point, melting point, molar heat capacity, electron affinity, and ionization energy are identified as the potential descriptors for σ. Using a data set of 124 compounds, a Gradient Boost Regression (GBR) model is developed, which has very small root-mean-square error (rmse) of 0.22 S/cm and a high coefficient of determination (R 2 ) of 0.98 for prediction of log-scaled σ. Utilizing the predicted σ values, τ el has been calculated for a wide range of temperatures. ML predicted τ el values outperform the τ def , obtained from the deformation potential model. The developed GBR model for accurate prediction of σ could accelerate the assessment of the efficiency of the thermoelectric materials with unprecedented accuracies.

From studying the atomic structure and chemical behavior to the discovery of new materials and investigating properties of existing materials, machine learning (ML) has been employed in realms that are arduous to probe experimentally. While numerous highly accurate models, specifically for property prediction, have been reported in the literature, there has been a lack of a generalized framework. Herein we propose a novel feature selection approach that enables the development of a unified ML model for property prediction for several classes of materials. It involves an ingenious blending of selected features from various classes of data such that the resultant feature set equips the model with global data descriptors capturing both class-specific as well as global traits. We took accurate band gaps of three distinct classes of 2D materials as our target property to develop the proposed feature blending approach. Using Gaussian process regression (GPR) with the blended features, the ML model developed here resulted in an average root-mean-squared error of 0.12 eV for unseen data belonging to any of the participating classes. The feature blending approach proposed here can be extended to additional classes of materials and also to predict other properties.