If your molecule has unpaired electrons, you should run an unrestricted calculation, in principle. However, if this exhibits convergence problems (or if you simply want to save time: an unrestricted calculation takes a factor 2 more CPU time and data storage), you may consider to do it in two steps. First, run a spin-restricted calculation. Then perform a spin-unrestricted calculation using the restricted TAPE21 as a restart file. In the follow-up calculation you should specify the precise occupation numbers for the state you're interested in, and use the SCF input key to specify only one SCF cycle (iterations=1). This prohibits convergence (so you keep the converged restricted orbitals) and gives you a fairly adequate approximation to a converged unrestricted result. See also the H2 example run for a discussion in the Examples document.
An unrestricted calculation does not necessarily yield the multiplet configuration (triple, doublet ...). This is a rather complicated matter, see the discussion on multiplet states, key SLATERDETERMINANTS.
