Why should primordial perturbations be in a vacuum state?

In order to calculate the power spectrum generated during a stage of inflation, we have to specify the quantum state of the inflaton perturbations, which is conventionally assumed to be the Bunch-Davies vacuum. We argue that this choice is justified only if the interactions of cosmological perturbations are strong enough to drive excited states toward the vacuum. We quantify this efficiency by calculating the decay probabilities of excited states to leading order in the slow-roll expansion in canonical single-field inflationary models. These probabilities are suppressed by a slow-roll parameter and the squared Planck mass, and enhanced by ultraviolet and infrared cut-offs. For natural choices of these scales decays are unlikely, and, hence, the choice of the Bunch-Davies vacuum as the state of the primordial perturbations does not appear to be warranted.