Dwarf galaxies are the most common galaxies in the universe and the least understood. Below about 10⁹ solar masses in halo mass, the relationship between dark matter and visible stars becomes wildly unpredictable. Two halos of identical mass can host stellar populations differing by a factor of 100. The scatter is not measurement noise — it reflects genuine diversity in formation histories, and reproducing it in simulations has been difficult.
Lin et al. (2602.22206) simulate eight dwarf galaxies with the same halo mass (~10⁹ solar masses) at extremely high resolution, following them from cosmological initial conditions to 1.2 billion years after the Big Bang — by which time reionization has expelled most of their gas. The stellar-to-halo mass ratio spans nearly two orders of magnitude, from 5×10⁻⁵ to 2×10⁻³. Same final halo mass. Vastly different stellar content.
The diversity traces to formation timing. Halos that assembled earlier — reaching higher masses before reionization — had more time to form stars in dense gas. Halos that assembled later started forming stars in gas that was already being heated by the ultraviolet background. The reionization epoch acts as a deadline: what you built before the UV background turned on determines what you end up with.
The most surprising result concerns compactness. Some simulated galaxies have stellar half-mass radii of about 30 parsecs — as small as observed ultra-compact dwarf galaxies (UCDs). UCDs are usually explained as the stripped nuclei of larger galaxies — their compactness attributed to tidal processes that removed the outer stars while preserving the dense core. The simulations show an alternative: UCDs can form in situ, in isolated dark matter halos, without any stripping.
The condition for compact formation is a threshold in central gas surface density. When the gas density exceeds about 30 solar masses per square parsec, star formation becomes highly efficient and concentrated. Below this threshold, star formation is spread out and produces diffuse stellar distributions. The threshold is analogous to the condition for forming dense, massive star clusters — the same physics that makes a globular cluster compact can make an entire dwarf galaxy compact, if the gas conditions are right.
This threshold picture explains the diversity more parsimoniously than invoking different physical mechanisms for different morphologies. The same halos, the same cosmological framework, the same star formation physics — the only difference is whether the gas reached the critical density before reionization shut everything down. Dense gas makes compact galaxies. Diffuse gas makes diffuse galaxies. The gas density at the moment of peak star formation is the determining variable.
The resolution requirement is severe. Capturing 30-parsec half-mass radii in cosmological simulations requires resolving the internal structure of forming star clusters within their parent halos. Most cosmological simulations don't have this resolution. The SIRIUS simulations achieve it by focusing computational resources on eight individual halos rather than statistical samples. The trade-off between resolution and sample size is real — you can simulate many objects poorly or few objects well. For understanding the physics of the diversity, few-but-well is the right choice.