Exploitation of the Concept of Vicariance to Predict the Space Group of Lithiated Manganese or Cobalt Oxides

Abstract

In this work, a machine learning program was used to predict the crystal structure of lithiated manganese or cobalt oxides based only on their chemical composition. The composition and crystal structure of lithiated iron oxides were used as trial matrix. To assign the crystal structure, the Euclidean distance between the stoichiometric coefficients of the elements of the compound under testing and the trial compound was calculated. The softmax function was used to convert this distance into a probability distribution. The compound under test was assigned the space group of the training compound that appeared with the highest percentage. The logarithmic cross-entropy loss was used in evaluating the forecast results. The results showed that the program, for logarithmic cross-entropy loss values between 0.2 and 0.3, can predict the crystalline group with an accuracy of about 0.67. In the same range, sensitivity and precision values are placed in a range between 0.6 and 0.8, respectively, and the F1_Score reaches values above 0.62.

Keywords

• crystallographic structure
• machine learning
• Li-ion battery
• Fe
• Mn
• Co