Software tools for data-driven research and their application to thermoelectrics materials discovery
1 like283 views
This document summarizes several software tools for materials data science and their application to thermoelectrics materials discovery. It discusses Atomate for high-throughput calculations, Matminer for materials feature extraction and machine learning, AMSET for electron transport modeling, and integration with the Materials Project database. Example applications are described like using order parameters for structure characterization and a computational screening identifying new thermoelectric materials like YCuTe2.
Software tools for data-driven research and their application to thermoelectrics materials discovery
- 1. MATERIAL FEATURES PROPERTY TiO2 rutile F11 F12 … F1N gap = 3.0 eV C diamond F21 F22 … F2N gap = 5.5 eV … … … … … … PbTe rocksalt FM1 FM2 … FMN gap = 0.3 eV Python ML Libraries Data Featurization Data Retrieval Data Visualization Materials Databases MPDSCitrine Materials Project Software tools Materials Project integration Software tools for data-driven research and their application to thermoelectrics materials discovery Anubhav Jain, Energy Storage & Distributed Resources Division, Berkeley Lab Thermoelectrics discovery Atomate is a library of standardized workflows for high-throughput computational materials science. One can provide a crystal structure and select from >15 different types of calculation procedures (e.g., band structure, elastic tensor, thermal expansion, etc.) to perform. Atomate scales to millions of calculations at large supercomputing centers and produces a database of organized results. www.github.com/hackingmaterials/atomate (production) Matminer retrieves data from the APIs of several large databases and assists the user with feature extraction, implementing over 20 different featurization classes that can g e n e r a t e t h o u s a n d s o f physically-relevant materials descriptors. Matminer includes a built-in visualization package and interfaces to standard Python-based libraries for machine learning. www.github.com/hackingmaterials/matminer (released) The AMSET project develops and implements an ab initio model for electron transport (mobility, Seebeck) that b a l a n c e s a c c u r a c y a n d computational efficiency. It achieves this by adapting classical scattering equations intended for single 1D parabolic bands to complex, anisotropic band structures c o m p u t e d f ro m d e n s i t y functional theory methods. www.github.com/hackingmaterials/amset (in development) The Materials Project produces computationally-generated m a t e r i a l s d a t a t h a t i s disseminated to over 45,000 registered users. The software tools developed in our project are integrated with the Materials Project. For example, the atomate software will be used by Materials Project to run millions of simulations in the upcoming year. We have recently developed a set of order parameters to characterize the local environment of a site in a structure. Order parameter functions produce a value of 1 when matching a target geometry (e.g., “tetrahedral”) and continuously fade to zero as the neighbor geometry deviates from the target. The set of order parameters are assembled into a “fingerprint” that characterizes a site’s resemblance to ~20 coordination motifs. The site fingerprints are then assembled into a structure fingerprint. One application deployed on MP is to compute “similarities” between crystal structures in terms of fingerprint distance, which is robust even under distortion or alloying. We used the atomate library to compute the electron transport properties of ~48,000 inorganic materials using the BoltzTraP method under a constant relaxation time approximation. This database (~300GB) is downloadable online through the Dryad repository and is used in our project as the basis for identifying potential new thermoelectric compositions. One novel thermoelectric composition identified from the computational screening is p- type YCuTe2, which exhibits favorable band convergence and exhibits a peak zT of ~0.75 (in-line with computational estimates). To our knowledge, this is the highest experimental zT from a composition first suggested from computational guidance. Another material, TmAgTe2, was suggested but its experimental zT (~0.35) was limited by doping issues. AMSET! >45,000 users! Target: W similar structures detected Cs3Sb! TiGaFeCo! CeMg2Cu! www.materialsproject.org/materials/mp-91/ 1.0E+02 1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08 200 300 400 500 600 700 800 900 1000 1100 Mobility(cm2/V*s) Temperature (K) IMP PIE ACD POP overall expt experiment computation We have also investigated traditionally under-explored chemistries such as phosphide thermoelectrics. Contrary to the belief that phosphides should have high thermal c o n d u c t i v i t y ( κ ) , o u r computations suggest that the m i n i m u m κ o f t h e s e compounds can be quite low. Experimental tests of cubic NiP2 confirms low κ (~1.0 W/ m*K) but with poor electronic properties. Funding Funding for this research was provided by the U.S. Department of Energy, Basic Energy Sciences, Materials Science Division through an Early Career Grant and through the Materials Project center grant EDCBEE. Computing resources were provided by the National Energy Research Scientific Computing Center. Registered users over time www.materialsproject.org Electron mobility of n-type GaAs (ne = 3.0 * 1013 cm-3)