Below 20 solar masses, they formed in pairs. Above 45, they were thrown together.
Binary black hole mergers detected by LIGO-Virgo-KAGRA form through two channels (arXiv:2603.20430): isolated binary evolution (two stars born together, evolving and dying as a pair) and dynamical interactions (black holes meeting in dense star clusters through gravitational encounters).
Note (2026-11-15; read against the paper's full text): the 20 and 45 solar-mass boundaries are outputs of the paper's semi-analytic population-synthesis models (B-POP), whose local merger rate of 17.5 to 24.1 per cubic gigaparsec per year the paper calls consistent with LVK limits; they are not an observed boundary. I did not find the phrase 'too heavy' in the paper; its statement is that the most massive detected events, GW190521 and GW231123, hardly reconcile with isolated stellar evolution, followed by 'but see' two references for the other view. The paper also looks at individual sources: it says only GW190521 falls in a region it identifies as favouring a high-generation primary and that GW231123 looks like an outlier of one distribution, while saying the complexity of the astrophysics can hinder definitive conclusions about origin.
Population synthesis with the B-POP code, incorporating stellar evolution and metallicity changes across cosmic time, reveals a mass boundary. Below ~20 solar masses, isolated binary evolution dominates — these black holes inherited their pairing from the stars that created them. Above ~45 solar masses, dynamical processes become significant — these massive black holes are too heavy to have formed through standard isolated channels and show distinctive spin and mass-ratio characteristics.
The intermediate range (20-45 solar masses) is the contested zone where both channels contribute. Individual events in this range can't be confidently assigned to either channel — the complexity of the underlying astrophysics prevents definitive conclusions about any single merger.
The structural insight: the formation channel leaves a signature in the population, not in the individual. You can't look at one merger and determine how it formed (except at the extremes). But the distribution of masses, spins, and mass ratios across the full population constrains the relative contribution of each channel. The population is the telescope — individual events are too noisy to resolve the question, but their aggregate statistics separate the channels. The signal is statistical, not singular.