Application of machine learning to frustrated magnets

Understanding complex phases in materials showing glassy, or highly correlated liquid states is extremely challenging. Conventional simulation approaches struggle to deal with the need to account for multiple and competing interactions, as well as relate models and data together. At the heart of the problem is the difficulty in extracting accurate models from experimental data. Further, the scale and complexity of data, such as from neutron scattering on frustrated magnets, has made any form of quantitative analysis very demanding. By training neural nets over large numbers of models, machine learning techniques discriminate between different models and identify different physical regimes including formation of spin liquids and unusual broken symmetries. The neural nets can for example learn diffuse scattering directly in three-dimensional reciprocal space and can extract the most relevant information, denoise, and remove background by projecting the experimental neutron data on a finite dimensional space that is determined by an autoencoder. Machine learning outputs the potential models, quantifies their uncertainty and identifies and classifies different regimes that could be reached by modifying or applying external forces/fields to the material under consideration. This approach is shown to provide better understanding of the formation of a glass on cooling the spin liquid Dy2Ti2O7, quantum spin liquids on honeycomb lattices, and understanding the interactions and phases in RuCl3. Examples of them as well as powder scattering data are shown.