Accurate simulation of magnetic materials using computational methods is essential for under-standing their fundamental behavior and enabling their use in technological applications. In this work, I use first-principles calculations to investigate systems with magnetic properties and to develop new methods for predicting the behavior of these materials. The systems studied are characterized by magnetic moments that are localized near the atomic sites. The paramagnetic state, at which these magnetic moments are disordered, and the magnetic order-disorder transition are of specific interest in this work.

To better capture finite-temperature magnetic behavior, a machine learning (ML) model is developed to predict the magnitudes of the magnetic moments at finite temperatures. This enables the inclusion of longitudinal spin fluctuations in coupled spin-lattice dynamics simulations, which would otherwise be computationally prohibitive. The ML model is applied to Fe at both the magnetic transition temperature, 1043 K, and at a pressure and temperature comparable to the conditions of the Earth’s inner core.

Evidently, the magnetic order-disorder transition temperature of ferromagnetic materials, known as the Curie temperature, is a fundamental property, since these materials lose their macroscopic magnetization above this point. Predicting this temperature is therefore crucial for the discovery and design of new magnetic materials. An approach is proposed which is based on the energy difference between magnetically ordered and disordered states, obtained from density functional theory (DFT) calculations. This method offers a balance between accuracy and computational efficiency, allowing its application to a wide variety of systems and making it suitable for high-throughput screening. The approach is fitted to and benchmarked against several known ferro- and ferrimagnetic materials and further evaluated on a particularly challenging class of systems: substitutionally disordered alloys. Finally, this approach enables a high-throughput exploration of Fe-, Mn-, and Co-containing systems to identify promising candidates for magnetic applications.

In addition, the debated role of constraining fields in DFT calculations for constrained non-collinear magnetism is investigated. The study shows that these fields can be used to propagate the transverse dynamics of magnetic moments, thereby providing a theoretical foundation for their use in adiabatic spin dynamics simulations.

In this licentiate thesis, the search for new materials is presented within the paradigm of materials informatics, which uses high-throughput, density-functional-theory-optimized workflows and machine learning to discover new materials from combinations of the elements of the periodic table. I developed workflows to investigate phase transitions in pseudo-binary fluoride perovskite solid solutions and to estimate the Curie temperature of magnetic materials. Modern material science offers a vast array of computational tools, ranging from machine learning and artificial intelligence to high-performance computing and advanced codes maintained by dedicated researchers. To leverage these techniques and perform large-scale theoretical investigations, workflows can be developed to manage the workload of executing hundreds of thousands of calculations efficiently.

In paper I, we developed a workflow to identify pseudo-binary solid solutions of fluoride perovskites that share at least one common atomic species apart from fluoride. The set of candidate endpoints investigated consists of 3,969 unique perovskites, which would yield 7,874,496 material systems if not systematically reduced through a screening process. The screening involves three steps: (i) verifying that the endpoints are non-conductive by calculating their band gaps using density functional theory (DFT), (ii) ensuring that the endpoints can form a solution with a phase transition along the composition interval, and (iii) assessing the alloy’s synthesizability by comparing it to known theoretical phases with similar stoichiometry. The screening process identified 111 promising solid solutions, and 11 were studied in detail to validate the initial predictions, showing good agreement.

In paper II, we developed a workflow that uses DFT calculations to estimate the Curie temperature of magnetic materials. The process consists of two steps: first, calculating the magnetic ground state, and second, constructing a supercell containing at least 12 magnetic atoms for a disordered local moment calculation. The resulting data, combined with parameters fitted to experimental results, enable prediction of the Curie temperature with a mean absolute error of approximately 126 K.

These works highlight the usefulness of automated computational workflows and the opportunities they create for investigating large numbers of materials and deriving meaningful conclusions from extensive datasets.