We investigate the peculiar velocities predicted for galaxy clusters by the
ories in the cold dark matter family. A widely used hypothesis identifies r
ich clusters with high peaks of a suitably smoothed version of the linear d
ensity fluctuation field. Their peculiar velocities are then obtained by ex
trapolating the similarly smoothed linear peculiar velocities at the positi
ons of these peaks. We test these ideas using large high-resolution N-body
simulations carried out within the Virgo supercomputing consortium. We find
that at early times the barycentre of the material that ends up in a rich
cluster is generally very close to a peak of the initial density field. Fur
thermore, the mean peculiar velocity of this material agrees well with the
linear value at the peak. The late-time growth of peculiar velocities is, h
owever, systematically underestimated by linear theory. At the time when cl
usters are identified, we find their rms peculiar velocity to be about 40 p
er cent larger than predicted. Non-linear effects are particularly importan
t in superclusters. These systematics must be borne in mind when using clus
ter peculiar velocities to estimate the parameter combination sigma(8)Omega
(0.6).