Provenance, workflows, and crystallographic tools in materials science: AiiDA, spglib, and seekpath
- PDF / 4,100,078 Bytes
- 7 Pages / 585 x 783 pts Page_size
- 112 Downloads / 195 Views
Introduction The growth of available storage, memory, and central processing unit (CPU) speed has arguably outpaced advances in the development of algorithms capable of taking advantage of an entire supercomputer for single computations, at least in common electronic structure and atomistic simulations in computational materials science. As a result, the trend in the field is shifting toward exploiting supercomputers to run large numbers of simulations, each taking only a relatively small amount of time and resources (e.g., plane-wave-basis density functional theory [DFT] codes typically running for a few days on a hundred cores). It has become increasingly possible, using thousands of parallel individual calculations, to rapidly scan wide parameter spaces, such as atomic structure and composition. This “screening” approach allows for the examination of many materials variations by computation of their properties, selection of promising areas to explore with more accurate methods and experiments, and the ultimate discovery of new materials with optimal properties.1 Similar trends, sometimes termed high-throughput computing, have emerged in diverse areas of computation and information science, and many parallels exist in challenges and solutions across disciplines.
The first ingredient required for large-scale screening efforts is automation, eliminating the time-consuming task of manually managing the life cycle of each calculation, from input generation and deployment to output retrieval. In the solid-state electronic-structure domain, various tools have appeared early on in the last decade to address the needs to automate, among others, DFT energy computations (e.g., AFLOW,2 Materials Project,3 OQMD,4) and crystal structure manipulation, such as the Python codes ASE5/ pymatgen,6 which perform many useful operations such as creating supercells, surfaces, systems with defects, and many more advanced features. Despite increasing capabilities, however, many available tools are not interoperable and mostly focus on either specific computational codes or a narrow set of computation types. The second required ingredient is the ability to maintain data quality, accessibility, and reproducibility. These challenges call for the development of new software infrastructures to couple automatic materials computations to database storage solutions. Fortunately, the availability of mature tools and concepts in the areas of databases and automation brings tremendous opportunities to data-intensive computational investigations of materials.
Giovanni Pizzi, École Polytechnique Fédérale de Lausanne, Switzerland; [email protected] Atsushi Togo, Department of Materials Science and Engineering, Kyoto University, Japan; [email protected] Boris Kozinsky, Harvard University, USA; [email protected] doi:10.1557/mrs.2018.203
696
• VOLUME 43 • SEPTEMBERUniversity 2018 • www.mrs.org/bulletin Downloaded MRS fromBULLETIN https://www.cambridge.org/core. of Western Ontario, on 10 Sep 2018 at 08:42:26, subject to the Cambridge Cor
Data Loading...
