Strain data augmentation enables machine learning of inorganic crystal geometry optimization

• Published in: Patterns (New York, N.Y.) N.Y.), 2023-02, Vol.4 (2), p.100663-100663, Article 100663
• Main Authors: Dinic, Filip, Wang, Zhibo, Neporozhnii, Ihor, Salim, Usama Bin, Bajpai, Rochan, Rajiv, Navneeth, Chavda, Vedant, Radhakrishnan, Vishal, Voznyy, Oleksandr
• Format: Article
• Language: English
• Subjects: bulk modulus; deep learning; force field; formation energy; geometry optimization; graph neural networks; materials discovery; strain
• ISSN: 2666-3899; 2666-3899
• DOI: 10.1016/j.patter.2022.100663

Machine-learning (ML) models offer the potential to rapidly evaluate the vast inorganic crystalline materials space to efficiently find materials with properties that meet the challenges of our time. Current ML models require optimized equilibrium structures to attain accurate predictions of formation energies. However, equilibrium structures are generally not known for new materials and must be obtained through computationally expensive optimization, bottlenecking ML-based material screening. A computationally efficient structure optimizer is therefore highly desirable. In this work, we present an ML model capable of predicting the crystal energy response to global strain by using available elasticity data to augment the dataset. The addition of global strains improves our model’s understanding of local strains too, significantly improving the accuracy of energy predictions on distorted structures. This allows us to construct an ML-based geometry optimizer, which we used for improving the predictions of formation energy for structures with perturbed atomic positions. •Trained model to predict strained and ground-state structures•Our enhanced model was able to predict both local and global deformations•Enhanced model enabled a machine-learning-based Monte-Carlo optimizer Currently, only a small quantity of known inorganic crystals have been computationally calculated or theoretically verified. Machine learning (ML) is a powerful tool that can be used to search through the large inorganic crystal space. However, ML predictions are dependent on the input of accurate structural information, which is not available for new materials. An ML-based optimizer can be used to obtain ground-state structural information; however, the model should be able to differentiate ground-state and high-energy structures. Here, we demonstrate a model that was trained on both ground-state and systematically distorted structures. This allows our enhanced model to understand the effects local and global distortions have on crystal energy. Furthermore, our model was implemented in an ML-based geometry optimizer for inorganic crystals. We show that our optimizer can be used to improve prediction of formation energy for structures with perturbed atomic positions. Predicting novel materials with machine learning (ML) requires accurate structural information. For novel materials, equilibrium structures are not known and must be obtained with computationally expensive ab initio methods. A