Every textbook on general relativity states that light propagates along null geodesics. Although there are many senses in which this is true at sufficiently-high frequencies, it breaks down more generally. Different notions of "propagation direction" also become distinct at lower frequencies.
This talk will focus on the motion "as a whole" of electromagnetic pulses with large (but not infinitely-large) frequencies.
Angular momentum then affects the motion, resulting in null but non-geodesic trajectories. Precise answers depend, however, on what exactly is meant by the "pulse as a whole:" its centroid. There are many centroid definitions which appear to be reasonable, but surprisingly, some of these appear to be nowhere near the pulse itself! This turns out to be an unphysical artifact of the high-frequency approximation.
Although massless spinning wavepackets appear generically in the usual approximations, no such pulses can exist non-perturbatively.
High-frequency approximations break standard features of Maxwell theory, such as the fact that exact electromagnetic stress-energy tensors satisfy positive-energy conditions.
It is very easy to miss this fact in practice, and doing so can result in qualitatively-incorrect conclusions regarding the motion and localization of electromagnetic waves.
Abraham Harte (Ireland): The localization and propagation of high-frequency electromagnetic waves
