Abstract
SummaryThe discovery of new inorganic materials in unexplored chemical spaces necessitates calculating total energy quickly and with sufficient accuracy. Machine learning models that provide such a capability for both ground-state (GS) and higher-energy structures would be instrumental in accelerated screening. Here, we demonstrate the importance of a balanced training dataset of GS and higher-energy structures to accurately predict total energies using a generic graph neural network architecture. Using 16,500 density functional theory calculations from the National Renewable Energy Laboratory (NREL) Materials Database and 11,000 calculations for hypothetical structures as our training database, we demonstrate that our model satisfactorily ranks the structures in the correct order of total energies for a given composition. Furthermore, we present a thorough error analysis to explain failure modes of the model, including both prediction outliers and occasional inconsistencies in the training data. By examining intermediate layers of the model, we analyze how the model represents learned structures and properties.
Highlights
With the advances in computing power and methodologies, computational chemistry and materials science have made great strides in accelerating discovery of molecules and materials with tailored properties.[1,2] The ability to perform large-scale ab initio calculations, in particular those based on density functional theory (DFT), has been instrumental in inorganic functional materials discovery.[3,4,5,6,7] computational searches have largely focused on known materials documented in crystallographic databases
We develop a graph neural network (GNN) built upon existing architectures to predict the total energy of ground-state (GS) as well as hypothetical higher-energy structures generated for structure prediction.[16]
The formation enthalpy (DHf) of a crystal structure with a chemical composition AxByCz can be calculated from the DFT total energy as, DHf = Etotal À xm0A À ym0B À zm0C, where Etotal is DFT total energy of AxByCz with DHf and Etotal expressed per formula unit and m0i are the reference chemical potentials of elements, typically under standard conditions
Summary
The discovery of new inorganic materials in unexplored chemical spaces necessitates calculating total energy quickly and with sufficient accuracy. Machine learning models that provide such a capability for both ground-state (GS) and higher-energy structures would be instrumental in accelerated screening. We demonstrate the importance of a balanced training dataset of GS and higher-energy structures to accurately predict total energies using a generic graph neural network architecture. Using $16,500 density functional theory calculations from the National Renewable Energy Laboratory (NREL) Materials Database and $11,000 calculations for hypothetical structures as our training database, we demonstrate that our model satisfactorily ranks the structures in the correct order of total energies for a given composition. By examining intermediate layers of the model, we analyze how the model represents learned structures and properties
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
