Skip to Content
You are currently on the new version of our website. Access the old version .
ComputationComputation
  • Article
  • Open Access

9 May 2023

15 Pages

A Two-Step Machine Learning Method for Predicting the Formation Energy of Ternary Compounds

,
,
,
,
,
and
1
Dynamics of Condensed Matter, Chair of Theoretical Chemistry, University of Paderborn, Warburger Str. 100, 33098 Paderborn, Germany
2
Department of Mathematics, Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany
3
Department of Computer Science, University of Paderborn, Warburger Str. 100, 33098 Paderborn, Germany
4
Paderborn Center for Parallel Computing (PC2), University of Paderborn, Warburger Str. 100, 33098 Paderborn, Germany
This article belongs to the Section Computational Chemistry

Abstract

Predicting the chemical stability of yet-to-be-discovered materials is an important aspect of the discovery and development of virtual materials. The conventional approach for computing the enthalpy of formation based on ab initio methods is time consuming and computationally demanding. In this regard, alternative machine learning approaches are proposed to predict the formation energies of different classes of materials with decent accuracy. In this paper, one such machine learning approach, a novel two-step method that predicts the formation energy of ternary compounds, is presented. In the first step, with a classifier, we determine the accuracy of heuristically calculated formation energies in order to increase the size of the training dataset for the second step. The second step is a regression model that predicts the formation energy of the ternary compounds. The first step leads to at least a 100% increase in the size of the dataset with respect to the data available in the Materials Project database. The results from the regression model match those from the existing state-of-the-art prediction models. In addition, we propose a slightly modified version of the Adam optimizer, namely centered Adam, and report the results from testing the centered Adam optimizer.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.