AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-21
ASTROLABE: A 3D Particle-Level Surrogate for Cosmology Simulation with Physical Flow Matching
Systematic N-body scans over dark matter theory space are prohibitively expensive, and existing machine-learning emulators are limited by low resolution or by only being able to emulate specific observables. Astrolabe is a flow-matching generative emulator that acts directly on the 6N phase-space coordinates of N particles, taking a COLA-based coordinate prior and enforcing momentum and mass conservation, so in principle any observable can be measured with no voxel-resolution ceiling. In the Lambda-CDM cases tested it produces realistic fields with accurate power spectra, halo mass functions, velocity profiles, and halo density profiles, generalizes out-of-distribution to dark-acoustic oscillation initial conditions, and can be fine-tuned to emulate a TNG50-3-dark simulation at a fraction of the cost.