A Dynamics Data Set for Spin Ice

Spin ice is a magnetic system whose constituents fluctuate collectively to produce emergent magnetic monopoles at low temperatures. These monopoles affect an equilibration which can take days to complete, despite microsecond spin-flip processes. In the interest of solving this mysterious long-time dynamical behavior in spin ice materials, we have generated a benchmark stochastic time series data set for both pyrochlore spin ice Dy₂Ti₂O₇ and artificial checkerboard spin ice. We present the results of our analysis which implements both traditional and data-driven methods. We further show that machine learning methods are capable of learning from these data sets by training a deterministic convolutional neural network that can statistically reproduce the stochastic data. We hope to provide a path for the study of frustrated and topological magnetic systems open to the data science community.