Massive stars can be found in wide (hundreds to thousands AU) binaries with other massive stars. We use $N$-body simulations to show that any bound cluster should always have approximately one massive wide binary: one will probably form if none are present initially; and probably only one will survive if more than one are present initially. Therefore any region that contains many massive wide binaries must have been composed of many individual subregions. Observations of Cyg OB2 show that the massive wide binary fraction is at least a half (38/74) which suggests that Cyg OB2 had at least 30 distinct massive star formation sites. This is further evidence that Cyg OB2 has always been a large, low-density association. That Cyg OB2 has a normal high-mass IMF for its total mass suggests that however massive stars form they 'randomly sample' the IMF (as the massive stars did not 'know' about each other).

Approximately 10 per cent of star clusters are found in pairs, known as binary clusters. We propose a mechanism for binary cluster formation; we use N-body simulations to show that velocity substructure in a single (even fairly smooth) region can cause binary clusters to form. This process is highly stochastic and it is not obvious from a region's initial conditions whether a binary will form and, if it does, which stars will end up in which cluster. We find the probability that a region will divide is mainly determined by its virial ratio, and a virial ratio above 'equilibrium' is generally necessary for binary formation. We also find that the mass ratio of the two clusters is strongly influenced by the initial degree of spatial substructure in the region.