Date of Degree


Document Type


Degree Name





Karl Sandeman

Committee Members

Radhika Barua

Sophia Suarez

Nicolas Giovambattista

David Schwab

Subject Categories

Condensed Matter Physics


materials informatics, magnetocaloric


The magnetocaloric effect, by which a magnetic material experiences a change in temperature due to an applied magnetic field, can be used for refrigeration. The corollary to the magnetocaloric effect -- known as the pyromagnetic effect -- is the phenomenon by which a magnetic material experiences a thermally-induced change in magnetization that can be used to harvest thermal energy. This dissertation has two main parts: one focusing on novel materials for energy harvesting; and another focusing on methods of materials discovery for refrigeration purposes. Thermomagnetic power generation (TMG) is the process by which magnetic flux, which comes from a temperature-driven change of magnetization, is converted into usable energy.

The first part of this dissertation investigates the ways in which magnetically hard materials, which have strong magnetic anisotropy and non-zero magnetic remanence, can be incorporated into TMG cycles in order to expand the area of M-H plane available for energy conversion. Two cases are considered: (i) hard ferrite magnets as the functional material for a two-quadrant TMG cycle; (ii) and a hard magnet applying a bias eld to a soft functional material, thus opening the second quadrant of the M-H plane. Experiments on commercially available hard ferrites reveal that these materials are not yet good TMG candidates, but hard magnets with higher thermal conductivity and a greater change of magnetization with temperature could outperform existing TMG materials. Computational results using Radia indicate that biasing a soft magnet with a hard magnet is essentially equivalent to shifting the M-H loop by an amount proportional to the field of the biasing magnet. In the case of polycrystalline gadolinium, the work output increases from 13.6 J/kg without a bias field to 33.6 J/kg with a bias field for the same temperature range, but experimental verification is needed.

High-throughput workflows, guided by machine learning and other statistical methods in material informatics, are essential to streamlining the process of screening new possible materials, making the experimental process as targeted and efficient as possible. The second part of this dissertation describes two approaches to materials screening based on simulated data. The first approach is a modification of the screening method of Bocarsly et al., wherein lattice deformation across the magnetic phase transition (magnetic deformation) was computed based on the difference between unit cell dimension lengths with and without magnetic interactions in models based on density function theory (DFT). The addition of a phenomenological parameter, calculated from the electronic density of states available from the Materials Project, can enhance the prediction of magnetic eld-induced entropy change, Sm, with nearly 60% better mean squared error than using magnetic deformation alone.

A method of scraping the Materials Project database for DFT results on magnetically ordered compounds using the pymatgen open-source Python library is presented. By analyzing the electronic density of states, computed unit cell volume and other parameters obtained through the Materials Project database for over 20,000 materials, a data subset is compiled to propose directions for magnetocaloric material research.

Finally, a computational approach is established to discern between compositions which undergo a magnetic first-order phase transition (FOPT) and second-order phase transition (SOPT) across two different doping schemes in the La-Fe-Si family of compounds. One potential categorization scheme uses the Frechet distance [7] as a metric for the similarity between the non-magnetic (NM) and ferromagnetic (FM) density of states (DOS). Another possible implementation considers the deviation of the difference between the NM and FM DOS upon resolving for the shift between the NM and FM DOS curves. The Frechet distance method provides a clear delineation between materials in the La-Fe-Si compound family which undergo a FOPT versus a SOPT across both doping schemes considered. When the majority-spin electronic DOS are appropriately scaled, the critical Frechet distance is near 0.77. These results present a promising new avenue of phase transition prediction based on DFT calculations.