Scholarlink

Abstract <p>Using machine learning programs, the prediction of 250 not yet obtained compounds of composition ABX (where A and B are different chemical elements, and X are As, Sn, Sb, Pb or Bi) with a crystal structure of the MgAgAs type was carried out and the values of their crystal lattice parameter were estimated. Using the cross-validation method, the best machine learning algorithms were selected for subsequent predicting. When making predicts about compounds that have not yet been synthesized, the most accurate programs were based on neural network training algorithms, support vector machines and <i>k</i>-nearest neighbors, for which the accuracy was determined to be 88.5, 91.0, and 88.4%, respectively. When predicting the value of the crystal lattice parameter of the predicted compounds, the best results were obtained using programs based on the Bayesian Ridge methods (coefficient of determination <i>R</i><sup>2</sup> = 0.959, mean absolute error MAE = 0.0370, mean square error MSE = 0.0030), ARD Regression (<i>R</i><sup>2</sup> = 0.950, MAE = 0.0401, MSE = 0.0036) and Ridge (<i>R</i><sup>2</sup> = 0.959, MAE = 0.0368, MSE = 0.0029), i.e., the deviation of the calculated values from the experimental ones was in the range of 0.0368 to 0.0401 A. When predicting new compounds and estimating their crystal lattice parameters, only the values of the properties of the chemical elements included in their composition were used.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of ABX Compounds (X = As, Sn, Sb, Pb, or Bi) with the MgAgAs Structure Type and Their Crystal Lattice Parameters

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • N. N. Kiseleva,
  • V. A. Dudarev,
  • A. V. Stolyarenko,
  • O. V. Senko,
  • A. A. Dokukin,
  • Yu. O. Kuznetsova

摘要

Abstract

Using machine learning programs, the prediction of 250 not yet obtained compounds of composition ABX (where A and B are different chemical elements, and X are As, Sn, Sb, Pb or Bi) with a crystal structure of the MgAgAs type was carried out and the values of their crystal lattice parameter were estimated. Using the cross-validation method, the best machine learning algorithms were selected for subsequent predicting. When making predicts about compounds that have not yet been synthesized, the most accurate programs were based on neural network training algorithms, support vector machines and k-nearest neighbors, for which the accuracy was determined to be 88.5, 91.0, and 88.4%, respectively. When predicting the value of the crystal lattice parameter of the predicted compounds, the best results were obtained using programs based on the Bayesian Ridge methods (coefficient of determination R2 = 0.959, mean absolute error MAE = 0.0370, mean square error MSE = 0.0030), ARD Regression (R2 = 0.950, MAE = 0.0401, MSE = 0.0036) and Ridge (R2 = 0.959, MAE = 0.0368, MSE = 0.0029), i.e., the deviation of the calculated values from the experimental ones was in the range of 0.0368 to 0.0401 A. When predicting new compounds and estimating their crystal lattice parameters, only the values of the properties of the chemical elements included in their composition were used.