Speaker
Description
We perform the highest-throughput search and evaluation of inorganic materials that can serve as excellent low-mass dark matter detectors. We use a graphical neural network (GNN) based on the GNNOpt architecture to learn and generate dielectric tensors, loss functions and sensitivity curves for both isotropic and anisotropic materials. Since the only required input for this GNN model is the structure of the unit cell, this analysis can now be expanded to any known inorganic material, increasing the number of materials analyzed by over 2 orders of magnitude. We apply this model to all materials available on the Materials Project database and show the results for absorption and scattering between dark matter and electrons. We also extend the GNN to predict anisotropy in the materials and calculate daily modulation in the interaction rates, which enables sensitivity to the dark matter wind. The training dataset is derived from nearly one thousand DFT-calculated dielectric tensors taken from Materials Project as well as previously calculated loss functions and sensitivity curves. The GNN predicts the output to percent level accuracy with no prior knowledge of DFT and the output obeys the Kramers-Kronig relations and physical sum rules similarly to the original dataset.