Data-driven design of inorganic materials with the Automatic Flow Framework for Materials Discovery
- PDF / 3,660,156 Bytes
- 6 Pages / 585 x 783 pts Page_size
- 107 Downloads / 206 Views
Introduction Density functional theory (DFT) implementations offer a reasonable compromise between cost and accuracy in ab initio materials science calculations,8 stimulating rapid development of automated frameworks and corresponding data repositories. Prominent examples include the Automatic Flow Framework for Materials Discovery (AFLOW),9–12 Novel Materials Discovery Laboratory (NOMAD),13 Materials Project,14 Open Quantum Materials Database (OQMD),15 Computational Materials Repository and its associated scripting interface Atomic Simulation Environment (ASE),16 and Automated Interactive Infrastructure and Database for Computational Science (AiiDA).17 Such repositories house an abundance of materials data. For instance, the AFLOW.org database contains more than 1.8 million compounds, each characterized by about 100 different properties.11,18–20 Investigations employing this data have not only led to advancements in modeling electronics,21–24 thermoelectrics,25,26 superalloys,27 and metallic glasses,28 but also the synthesis of new materials: for example, two new magnets, Co2MnTi and Mn2PtPd, which are the first discovered by computational approaches.29 Further advancements and discoveries are contingent on continued development and expansion of these materials repositories. New entries are generated both by (1) calculating the properties of previously observed compounds from 1–7
sources such as the Inorganic Crystal Structure Database,30 and (2) decorating structural prototypes31—populating crystal sites of existing structures with atoms of different elements— to predict new materials. Accurate computation of materials properties—including electronic, magnetic, chemical, crystallographic, thermomechanical, and thermodynamic features— demands a combination of reliable calculation parameters/ thresholds11 and robust algorithms that scale with the size/ diversity of the database. For example, convenient definitions for the primitive cell representation9 and high-symmetry Brillouin zone path10 have optimized and standardized electronic structure calculations. Careful treatment of spatial tolerance and proper validation schemes have finally enabled accurate and fully autonomous determination of the complete symmetry profile of crystals,32 which is essential for elasticity33 and phonon9,34–36 calculations. Beyond descriptions of simple crystals, exploration of complex properties33,37 and materials28,38 typically warrants advanced (and expensive) characterization techniques.39–41 Fortunately, state-of-the-art workflows33,37,38 and careful descriptor development28 have enabled experimentally validated modeling within a DFT framework. The combination of plentiful and diverse materials data11,18–20 and its programmatic accessibility19,20 justify the application of data-mining techniques. These methods can quantitatively resolve subtle
Corey Oses, Department of Mechanical Engineering and Materials Science, Duke University, USA; [email protected] Cormac Toher, Department of Mechanical Engineering and Materials Science, D
Data Loading...
