Reconstruction of Dark Matter and Baryon Density From GalaxiesConfirmed

by Jordan Krywonos (Perimeter Institute for Theoretical Physics)

Monday Jan 19, 2026, 2:30 PM → 3:00 PM America/Toronto

PI/4-400 - Space Room (Perimeter Institute for Theoretical Physics)

In cosmology, we often need to infer unobserved fields - such as dark matter and nonluminous baryons - from observed luminous tracers like galaxies. Traditional reconstruction techniques typically rely on the halo-model, which assumes radially symmetric density profiles centered on each tracer. These assumptions can limit their accuracy, particularly on nonlinear scales, and usually only use one galaxy property to inform the underlying matter distributions. I will present a novel machine learning approach that uses a hybrid graph neural network–convolutional neural network (GNN–CNN) architecture to map discrete galaxy catalogs to continuous fields. Using CAMELS simulations as ground truth, I will benchmark this method against traditional reconstruction techniques. I will demonstrate that the GNN-CNN yields the best reconstruction and can improve cross-correlation analyses such as kinetic Sunyaev Zel’dovich velocity reconstruction.