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 Emerging machine-learned models have enabled efficient and accurate prediction of compound formation energy, with the most prevalent models relying on graph structures for representing crystalline materials. Here, we introduce an alternative approach based on sparse voxel images of crystals. By developing a sophisticated network architecture, we showcase the ability to learn the underlying features of structural and chemical arrangements in inorganic compounds from visual image representations, subsequently correlating these features with the compounds’ formation energy. Our model achieves accurate formation energy prediction by utilizing skip connections in a deep convolutional network and incorporating augmentation of rotated crystal samples during training, performing on par with state-of-the-art methods. By adopting visual images as an alternative representation for crystal compounds and harnessing the capabilities of deep convolutional networks, this study extends the frontier of machine learning for accelerated materials discovery and optimization. In a comprehensive evaluation, we analyse the predicted convex hulls for 3115 binary systems and introduce error metrics beyond formation energy error. This evaluation offers valuable insights into the impact of formation energy error on the performance of the predicted convex hulls. 

Abstract Emerging machine-learned models have enabled efficient and accurate prediction of compound formation energy. While the prevalent models rely on graph structures for representing crystalline materials, we introduce an alternative approach using voxel images of crystals. By designing a deep and complex convolutional network, we demonstrate the capability to learn the underlying features of structural and chemical arrangements in inorganic compounds from this visual image representation and map them to the compounds’ formation energy. Our model achieves accurate formation energy prediction by utilizing skip connections in a deep convolutional network and incorporating augmentation of rotated crystal samples during training, performing on par with state-of-the-art methods. By adopting visual images as an alternative representation for crystal compounds and harnessing the capabilities of deep convolutional networks, this study extends the frontier of machine learning for accelerated materials discovery and optimization. In a comprehensive evaluation, we analyze the predicted convex hulls for 3,115 binary systems and introduce error metrics beyond formation energy error. This evaluation offers valuable insights into the impact of formation energy error on the performance of the predicted convex hulls. 

We present the successful synthesis and characterization of a one-dimensional high-entropy oxide (1D-HEO) exhibiting nanoribbon morphology. These 1D-HEO nanoribbons exhibit high structural stability at elevated temperatures (to 1000°C), elevated pressures (to 12 gigapascals), and long exposure to harsh acid or base chemical environments. Moreover, they exhibit notable mechanical properties, with an excellent modulus of resilience reaching 40 megajoules per cubic meter. High-pressure experiments reveal an intriguing transformation of the 1D-HEO nanoribbons from orthorhombic to cubic structures at 15 gigapascals followed by the formation of fully amorphous HEOs above 30 gigapascals, which are recoverable to ambient conditions. These transformations introduce additional entropy (structural disorder) besides configurational entropy. This finding offers a way to create low-dimensional, resilient, and high-entropy materials. 

A simple device for heating single-crystal samples to temperatures ≥2000 °C in ultrahigh vacuum that is compatible with the standard sample plates used in a common commercial scanning tunneling microscope (STM) is described. Heating high melting point samples to higher temperatures than is possible with many existing STM sample holders is necessary to obtain clean, well-ordered surfaces. Results are demonstrated for the (0001) surface of ZrB2, which has a melting point of 3050 °C. 

We report the P-V-T equation of state measurements of B4C to 50GPa and approximately 2500K in laser-heated diamond anvil cells. We obtain an ambient temperature, third-order Birch–Murnaghan fit to the P-V data that yields a bulk modulus K0 of 221(2) GPa and derivative, (dK/dP)0 of 3.3(1). These were used in fits with both a Mie–Grüneisen– Debye model and a temperature-dependent, Birch– Murnaghan equation of state that includes thermal pressure estimated by thermal expansion (α) and a temperature-dependent bulk modulus (dK0/dT). The ambient pressure thermal expansion coefficient (α0+α1T), Grüneisen γ (V)=γ 0(V/V0)q and volumedependent Debye temperature, were used as input parameters for these fits and found to be sufficient to describe the data in the whole P-T range of this study.