6-3. AFLOW: integrated infrastructure for autonomous computational materials discovery

Cormac Henry Toher

Duke University, USA

Abstract: The AFLOW framework for computational materials design integrates modules to generate new materials structures using crystallographic prototypes with routines to analyze symmetry and thermodynamic stability, calculate thermo-mechanical properties and perform automated error correction of DFT calculations. The data generated using AFLOW are available online at aflow.org and are programmatically accessible via the AFLOW REST-API and AFLUX Search-API.

Analysis of the dependence of formation enthalpy on the number of species using data extracted from the AFLOW data repository shows that the gain in formation entropy exceeds the gain in formation enthalpy as more species are added to the mixture, demonstrating the importance of disorder in the formation of multicomponent materials. Disordered materials are difficult to model directly using first principles calculations. Therefore, descriptors and thermodynamic models have been developed using data from ensembles of ordered calculations in the AFLOW data repository to predict the synthesizability of new disordered materials such as metallic glasses and high entropy carbides using the thermodynamic density of states, as well as to predict the transition temperatures and miscibility gaps for high entropy alloys. AFLOW data is also being employed to train machine-learning models to predict properties ranging from the electronic band gap to thermo-mechanical properties such as the bulk modulus and Debye temperature. These machine-learning models are now available for use via interactive online applications on the aflow.org web portal at aflow.org/aflow-ml, and are programmatically accessible via the new AFLOW-ML API.