Towards data-driven magnetic materials discovery
Citation:
ZIC, MARIO, Towards data-driven magnetic materials discovery, Towards Data-Driven Magnetic Materials Discovery, Trinity College Dublin.School of Physics.PHYSICS, 2017Download Item:
Abstract:
Magnetic materials underpin many of the technologies that define the world we live in. Despite the tremendous technological progress, the discovery of new magnetic materials has been rather slow. In this Thesis we explore and develop new, data-oriented, methods for the accelerated discovery and development of new magnetic materials. We utilize available theoretical databases of Heusler alloy properties to: develop a high-throughput (HT) screening procedure for the discovery of new permanent magnets, identify the defects in Mn-Ru-Ga thin films, and build machine learning (ML) models for predicting the structural and the magnetic properties of Heusler alloys. We identify a dozen materials, which meet all of the criteria for permanent magnet applications. The analysis of the HT data allowed us to understand the ideal composition of hard magnetic materials in the family of regular Heusler alloys, and to show how structure-property constraints affect the abundance of the potential candidates for technological applications. We find that hard permanent magnets occur with a frequency smaller than 1 in 10 000, with respect to the overall population of the regular Heusler alloys in the database. We then demonstrate that the ML techniques can be used both to improve the efficiency of the HT procedure and to perform a data-driven investigation of the material properties. In the case of the Mn-Ru-Ga thin films we show how the HT data can be utilized to guide the modeling of technologically relevant materials. The HT data was used to identify the nature of the defects that occur in the films, and hence, to obtain an accurate theoretical description of the material properties. We build ML models to investigate the magnetism of Fe in Heusler alloys. We then study how the local chemical environment affects its magnetic moment and thus address the structure-property relationship directly. We also show that new knowledge about the physics of materials can be extracted directly from the data. This work clearly demonstrates the potential that ML techniques have to offer in the analysis of a vast amount of materials data and paves the way for the future data-driven studies of magnetism.
Sponsor
Grant Number
Science Foundation Ireland (SFI)
SFI/HEA Irish Centre for High-End Computing (ICHEC)
European Union FP7 programme "ROMEO"
Trinity Centre for High Performance Computing (TCHPC)
Author's Homepage:
http://people.tcd.ie/zicmDescription:
APPROVED
Author: ZIC, MARIO
Sponsor:
Science Foundation Ireland (SFI)SFI/HEA Irish Centre for High-End Computing (ICHEC)
European Union FP7 programme "ROMEO"
Trinity Centre for High Performance Computing (TCHPC)
Other Titles:
Towards Data-Driven Magnetic Materials DiscoveryAdvisor:
Sanvito, StefanoPublisher:
Trinity College Dublin. School of Physics. Discipline of PhysicsType of material:
ThesisAvailability:
Full text availableKeywords:
Machine Learning, Magnetism, Computational MethodsMetadata
Show full item recordLicences: