Crystal Fractional Graph Neural Network for Energy Prediction of High-Entropy Alloys
Researchers Takanori Kotama and Yang Huang have proposed a novel Crystal Fractional Graph Neural Network designed to predict the energy of high-entropy alloys (HEAs). HEAs are gaining attention for their superior mechanical and thermal properties derived from complex atomic structures. The new model explicitly integrates local atomic environments with global compositional information through three key components: a crystal graph neural network using graph attention layers for local interactions, a fractional neural network for embedding global element fractions, and a feature fusion network to predict total crystal energy. Trained on 1,049 crystal structures and validated on 198 quaternary structures with hyperparameters optimized via Optuna, the model achieves root mean square error (RMSE) levels comparable to first-principles calculations. It maintains high accuracy even for low-energy configurations. However, the study acknowledges current limitations in handling large crystal cells, which the authors plan to address in future work to expand the model's applicability to more complex systems. This development represents a significant advancement in computational physics and materials science, leveraging artificial intelligence to accelerate the discovery and analysis of advanced alloy materials.
Wire timeline
Crystal Fractional Graph Neural Network for Energy Prediction of High-Entropy Alloys
Researchers Takanori Kotama and Yang Huang have proposed a novel Crystal Fractional Graph Neural Network designed to predict the energy of high-entropy alloys (HEAs). HEAs are gaining attention for their superior mechanical and thermal properties derived from complex atomic structures. The new model explicitly integrates local atomic environments with global compositional information through three key components: a crystal graph neural network using graph attention layers for local interactions, a fractional neural network for embedding global element fractions, and a feature fusion network to predict total crystal energy. Trained on 1,049 crystal structures and validated on 198 quaternary structures with hyperparameters optimized via Optuna, the model achieves root mean square error (RMSE) levels comparable to first-principles calculations. It maintains high accuracy even for low-energy configurations. However, the study acknowledges current limitations in handling large crystal cells, which the authors plan to address in future work to expand the model's applicability to more complex systems. This development represents a significant advancement in computational physics and materials science, leveraging artificial intelligence to accelerate the discovery and analysis of advanced alloy materials.
cs.AI updates on arXiv.org