Search for: All records

Abstract Machine learning models are susceptible to being misled by biases in training data that emphasize incidental correlations over the intended learning task. In this study, we demonstrate the impact of data bias on the performance of a machine learning model designed to predict the likelihood of synthesizability of crystal compounds. The model performs a binary classification on labeled crystal samples. Despite using the same architecture for the machine learning model, we showcase how the model’s learning and prediction behavior differs once trained on distinct data. We use two data sets for illustration: a mixed-source data set that integrates experimental and computational crystal samples and a single-source data set consisting of data exclusively from one computational database. We present simple procedures to detect data bias and to evaluate its effect on the model’s performance and generalization. This study reveals how inconsistent, unbalanced data can propagate bias, undermining real-world applicability even for advanced machine learning techniques. 

Abstract A metallic, covalently bonded carbon allotrope is predicted via first principles calculations. It is composed of ansp³carbon framework that acts as a diamond anvil cell by constraining the distance between parallelcis‐polyacetylene chains. The distance between thesesp²carbon atoms renders the phase metallic, and yields two well‐nested nearly parallel bands that cross the Fermi level. Calculations show this phase is a conventional superconductor, with the motions of thesp²carbons being key contributors to the electron–phonon coupling. Thesp³carbon atoms impart superior mechanical properties, with a predicted Vickers hardness of 48 GPa. This phase, metastable at ambient conditions, could be made by on‐surface polymerization of graphene nanoribbons, followed by pressurization of the resulting 2D sheets. A family of multifunctional materials with tunable superconducting and mechanical properties could be derived from this phase by varying thesp²versussp³carbon content, and by doping. 

First PXRD assisted crystal structure prediction method that can correct for temperature, strain, and choice of computational method. 

Metastable materials are abundant in nature and technology, showcasing remarkable properties that inspire innovative materials design. However, traditional crystal structure prediction methods, which rely solely on energetic factors to determine a structure’s fitness, are not suitable for predicting the vast number of potentially synthesizable phases that represent a local minimum corresponding to a state in thermodynamic equilibrium. Here, we present a new approach for the prediction of metastable phases with specific structural features and interface this method with the XTALOPT evolutionary algorithm. Our method relies on structural features that include the local crystalline order (e.g, the coordination number or chemical environment), and symmetry (e.g, Bravais lattice and space group) to filter the breeding pool of an evolutionary crystal structure search. The effectiveness of this approach is benchmarked on three known metastable systems: XeN8, with a two-dimensional polymeric nitrogen sublattice, brookite TiO2, and a high pressure BaH4 phase, which was recently characterized. Additionally, a newly predicted metastable melaminate salt, P1̅ WC3N6, was found to possess an energy that is lower than that of two phases proposed in a recent computational study. The method presented here could help in identifying the structures of compounds that have already been synthesized, and in developing new synthesis targets with desired properties.