Path measures for stochastic galaxy formation on layered halo graphs
Phys. Rev. D 114, 063043 – Published 22 September, 2026
DOI: https://doi.org/10.1103/nmd6-zkc1
Abstract
We introduce a graph path likelihood model (GPLM) for galaxy assembly histories on layered halo graphs extracted from hydrodynamic simulations. The graph contains temporal edges that encode progenitor-to-descendant transport, while coeval host edges encode environmental conditioning. On each graph, GPLM combines deterministic merger transport with learned residual drift and diffusion to assign probability weights to galaxy trajectories. This gives a path measure for coarse galaxy formation histories and extends the statistical description beyond end point galaxy properties. As a first realization, we train a graph neural likelihood model for stellar and gas mass assembly histories and show that it reproduces their main statistics while capturing environmentally conditioned fluctuations. We then use the learned path measure for calculations on fixed graphs, including probabilities for galaxies deficient in dark matter, controlled deformations of the gas response, and nonequilibrium diagnostics of environmentally dependent evolution. The present implementation uses fixed graphs, two fields, and a Gaussian residual sector, yet it demonstrates how graph structure, stochastic baryonic evolution, and response calculations can be organized within one probabilistic framework.
