How To Build The Universe in a Computer

Galaxy collisions can be predicted with striking confidence because gravity and fluid-like gas dynamics can be computed repeatedly over billions of years—turning a chaotic cosmic process into something simulatable. The core takeaway is that modern astrophysics doesn’t just “run models”; it builds practical approximations of an enormous many-body universe by breaking the problem into manageable steps and using algorithms that avoid brute-force calculations. That’s why events like the Milky Way and Andromeda’s eventual merger—complete with tidal streams, supernova-triggering gas compression, and the loss of spiral structure—can be treated as effectively inevitable outcomes of known physics.

The path to that capability runs through a key historical insight: exact solutions exist for only the simplest gravitational setups. Newton’s equations can be solved analytically for two bodies, but once three or more objects interact, the “3-body problem” has no tidy master equations. Erik Holmberg’s 1941 workaround used 37 light bulbs as gravity sources, iteratively measuring the combined “pull” at each bulb, updating velocities via Newton’s laws, and then letting the system evolve step by step—an early numerical calculation that foreshadowed today’s N-body simulations.

In an N-body simulation, the universe is represented by many particles whose motions follow gravity across discrete time steps. The method can be made arbitrarily accurate by shrinking the steps, and it already works well for smaller systems like the solar system. But galaxies are the bottleneck: a naive approach requires computing the force between every particle pair, scaling like N². For a million-particle star cluster, that means roughly a trillion interactions per time step—before even considering the need to model both fine detail (individual structures) and vast cosmic scales (large volumes).

Astrophysicists therefore rely on approximations that exploit how gravity weakens with distance. Tree codes reduce far-field interactions by grouping distant particles into hierarchical cubes, cutting the workload from N² down to about N log N. Particle-mesh methods convert particles into a density field on a grid, solve the gravitational potential efficiently using Fourier transforms, and use adaptive meshes to refine resolution where matter is dense. For gas dynamics, simulations shift from pure gravity to fluid equations: smoothed-particle hydrodynamics (SPH) replaces a rigid grid with moving tracer particles that carry the fluid’s properties. Modern runs often blend methods—SPH for large-scale gas flows paired with direct particle-particle N-body calculations for short-range accuracy.

These techniques enable increasingly ambitious virtual universes, from star formation in collapsing gas clouds to cosmological simulations spanning billions of light-years. The Millennium simulation modeled a 13-billion-light-year cube with over 300 billion particles, while AbacusSummit pushed to 70 trillion particles on supercomputers. Still, there’s a hard ceiling: typical simulations track only particle positions and velocities, not the full quantum information content of reality. A complete quantum universe would likely require a cosmically large quantum computer—so the practical limit is less about computing speed and more about what physics can be represented. The payoff remains real: simulations have become a primary way to test how structure forms, how galaxies evolve, and why the universe’s large-scale patterns emerge from microscopic rules.