MLatticeABC

MLatticeABC predicts lattice constants (unit cell edge lengths a, b, c and plane angles) of crystal materials to support crystal structure prediction and materials property prediction.

Key Features:

• Machine Learning Model: MLatticeABC uses a random forest machine learning model for lattice parameter prediction.
• Descriptor Set: It employs a descriptor set specifically tailored for predicting lattice unit cell edge lengths (a, b, c).
• Performance Metrics: It achieves R² = 0.973 for parameter a in cubic crystals, an average R² = 0.80 for a across all crystal systems, and R² values of 0.498–0.757 for parameters b and c.
• Generic Applicability: The method is applicable across materials with varied compositions, addressing limitations of prior models trained on small datasets.
• Lattice Angle Prediction: The approach shows improved performance for predicting lattice (plane) angles.

Scientific Applications:

• Crystal Structure Prediction: Provides predicted lattice parameters to inform crystal structure determination and screening.
• Materials Property Prediction: Enables prediction of material properties that depend on lattice parameters.

Methodology:

Training of a random forest model using a comprehensive descriptor set tailored to lattice unit cell edge lengths (a, b, c).