Convolutional Neural Networks to Assist the Assessment of Lattice Parameters from X-ray Powder Diffraction

Area of Science:

• Crystallography and Materials Science
• Computational Chemistry and Machine Learning

Background:

• Accurate determination of lattice parameters is crucial for understanding and predicting material properties.
• Traditional methods for lattice parameter estimation can be time-consuming and computationally intensive.

Purpose of the Study:

• To develop and evaluate convolutional neural networks (CNNs) for the automated estimation of lattice parameters in organic compounds.
• To investigate the impact of incorporating atomic composition data into CNN models for improved accuracy.

Main Methods:

• Trained two CNN architectures (XRD-CNN and XRDElem-CNN) on a large dataset of 92,085 organic compounds.
• Generated simulated X-ray diffraction (XRD) patterns for training and testing.
• XRDElem-CNN utilized diffraction patterns and a binary representation of unit cell atoms as input.

Main Results:

• XRD-CNN achieved moderate accuracy (MAPE: 11.04% for vectors, 7.40% for angles, 26.83% for volume).
• XRDElem-CNN demonstrated significantly improved accuracy (MAPE: 4.73% for vectors, 6.49% for angles, 6.05% for volume).
• XRDElem-CNN performance was validated using real XRD data and the Lp-search method.

Conclusions:

• CNNs, particularly XRDElem-CNN incorporating atomic information, offer a powerful and efficient approach for lattice parameter estimation.
• This method shows promise for high-throughput materials discovery and characterization.
• The proposed atomic representations are computationally efficient for CNN evaluation.