Machine Learning-Directed Discovery of Ultrahigh Young’s Modulus Crystals

Materials with ultrahigh Young’s modulus are essential
for
advanced applications such as aerospace components and energy devices.
Efficiently identifying the maximum Young’s modulus of materials
is a key factor in accelerating the development of advanced high-stiffness
materials. In this study, we developed a model based on a crystal
graph convolutional neural network, incorporating data processing
and model optimization techniques. These approaches effectively address
data scarcity and imbalance in high-modulus crystals while significantly
enhancing prediction accuracy, reducing the mean absolute error to
30.6 GPa on the test set. Guided by the model and validated through
first-principles calculations, we conducted high-throughput screening
on a comprehensive data set of over 1.16 million crystals. As a result,
we identified 31 ultrahigh Young’s modulus crystals, all exceeding
1000 GPa. Among them, OsO exhibited an exceptional modulus of 1557.9
GPa, making it one of the highest Young’s modulus crystals
discovered through machine learning-guided screening. Furthermore,
most of these crystals have formation energies below 0.4 eV/atom,
indicating favorable thermodynamic stability and potential experimental
synthesizability, making them promising candidates for advanced structural
applications. This study presents an effective approach for the discovery
of ultrahigh Young’s modulus crystals. The proposed method
significantly enhances the accuracy and efficiency of maximum Young’s
modulus identification, providing a valuable strategy for accelerating
the development of next-generation advanced materials for structural
and engineering applications.