Design of Novel Magnetic Materials with Machine-Learning and High-Throughput Techniques

Abstract

The importance of magnetic materials in modern applications is unquestionable. They are used in a wide range of applications, going from data storage devices to green energy production. This class of materials is well known since antiquity; yet the number of materials known to be magnetic has been limited to ~ 5000. A magnetic material is considered to be useful for applications if the Curie temperature (the temperature at which a ferromagnetic material ceases to be magnetic) is well above room temperature. If we enforce this constraint on the known materials then the number of useful ones reduces dramatically, this implies the need for the design of novel magnetic materials.

Traditionally, the search for suitable candidates is done experimentally. This task is not only time-consuming, reducing the throughput so much that it is impossible to explore the vast chemical spaces available, but is also expensive both in workforce and in resources.

Advances in computational materials science and artificial intelligence make the theoretical discovery of magnetic materials increasingly accessible. Tools like machine learning and high-throughput ab-initio calculations give us the ability to scan large chemical spaces in search of novel compounds that exhibit desired properties in a fraction of time. Moreover, the stability of these compounds can be assessed with high confidence. Suitable candidates can be studied with state-of-the-art ab initio simulations, and their properties can be predicted before they are synthesized experimentally.

This thesis is divided into 3 parts. In the first two chapters, we introduce our workflow for material exploration. This uses a combination of machine-learning and ab-initio methods for the search of thermodynamically stable ternary alloy materials. To achieve the needed throughput, we diminish the use of expensive ab-initio simulations by utilising Machine Learning Interatomic Potentials, MLIAPs. Using them as energy predictors, the exploration of large materials spaces becomes reachable. Then a methodology is explored for the creation of possible ternary candidates. Providing us with viable material candidates as a starting point is crucial for materials exploration workflows.

In the third chapter of this thesis, we will solely use high-throughput ab-initio methods to explore the Heusler family of materials. Heusler ternaries are well known for their magnetic properties, and interestingly, several high-performance magnets are discovered among them. Thoroughly, we use density functional theory, DFT, to search for antiferromagnetic and tetragonal distorted materials. The former are known to be used for spintronics applications, and the latter exhibit large magnetic crystalline anisotropy.

In the first part of this thesis, the reader will be introduced to the scope underpinning this work and the computational tools used. Then the main part of the work is divided into three chapters where the results of each project will be presented and discussed. The last part is focused on discussing the main findings of this work and presenting the future outlook.