Calibrating DFT Formation Enthalpy Calculations by Multifidelity Machine Learning

Abstract

The application of machine learning to predict materials properties measured by experiments are valuable yet difficult due to the limited amount of experimental data. In this work, we use a multifidelity random forest model to learn the experimental formation enthalpy of materials with prediction accuracy higher than the Perdew−Burke−Ernzerhof (PBE) functional with linear correction, PBEsol, and metageneralized gradient approximation (meta-GGA) functionals (SCAN and r 2 SCAN), and it outperforms the hotly studied deep neural network-based representation learning and transfer learning. We then use the model to calibrate the DFT formation enthalpy in the Materials Project database and discover materials with underestimated stability. The multifidelity model is also used as a data-mining approach to find how DFT deviates from experiments by explaining the model output.

Optimal thermodynamic conditions to minimize kinetic byproducts in aqueous materials synthesis

“…We note that the magnitude of the thermodynamic competition scale (tens of meV/atom) in aqueous solution-based synthesis is smaller than the thermodynamic limit for the synthesis of metastable inorganic materials (more than one hundred meV/atom) [11], and might be comparable with DFT errors [38,39,40].…”

Discovery of thermosetting polymers with low hygroscopicity, low thermal expansivity, and high modulus by machine learning

“…Recently, multi-fidelity learning has been successfully applied to chemistry and materials science. Several cases, including the prediction of formation enthalpy, 52 bandgap, 53–55 and energies, 56,57 are available in the literature.…”

In Silico High-Throughput Design and Prediction of Structural and Electronic Properties of Low-Dimensional Metal–Organic Frameworks

“…On the other hand, other relevant properties need to be calculated, such as performance on adsorption/storage of common gas molecules, partial atomic charges normally used to interpret trends while modeling chemical reactions, and density of states/band structures, which reveal detailed charge transport pathways. Machine learning techniques have also shown great promise in materials science research for the prediction of formation energies, adsorption energies, , band gap values, and designing new materials . The created crystal structures and their calculated property data gathered in our EC-MOF/Phase-I database provide an ideal data set for applying various machine learning techniques in order to explore their potentials for different applications.…”