Before proposing an answer, consider two types of face growth. Previous studies of snow-crystal growth have focused on the "normal growth process"; that is, the growth of basal and prism faces normal (perpendicular) to the plane of their surface. This process advances the face, and thus normal growth generally determines the primary habit (e.g., tabular vs columnar) as well as many features of the secondary habit (e.g., branching, hollowing). However, a crystal face also grows in area laterally. This lateral growth is usually outward, and its consequence is particularly clear after small facets first form on a spherical "droxtal". The facets form near the face centers and then spread outward laterally to transform the initially spherical crystal into a fully facetted polyhedral form. After the droxtal transforms to a sharply facetted polyhedral form, generally the normal growth process takes over. This process involves the generation of new steps from the outside perimeter that then travel inwards, towards the face center (e.g., Nelson and Baker, 1996). Due to the steps travelling inward, under some conditions, a central hollow can form in the face center. We refer to these features as 'center hollows'. If the hollow closes up again, the result is a 'center pocket'. Such features are very common in snow crystals, the hollow feature in particular having important atmospheric radiative effects.
Regarding the molecular-scale processes, both normal and lateral growth involve vapor-deposited molecules on a given face that subsequently migrate along the same face until finding a growth site. However, lateral growth differs in that it involves the surface migration of molecules over the edge of that face to the adjoining region.
Concerning the formation mechanism of corner and edge pockets, we argue that these pockets, unlike the center hollow, form due to a type of lateral growth. In particular, the lateral growth leads to overhanging protrusions in which the face 'overshoots' the adjoining rounded crystal orientation. Earlier, Akira Yamashita of Osaka Kyoiku University proposed such a growth mechanism, coining the term 'protruding growth' and used it to explain other types of pockets and features on snow crystals (Yamashita, 2016). In the case of our corner pockets, protruding growth occurs from the basal and two prism faces where they approach each other at a corner. When they meet at the corner, the air region under the overhangs becomes enclosed, thus forming a pocket. In the case of the edge pockets, the same occurs except from the basal and one prism face. Upon examination of historical snow-crystal images (Bentley and Humphreys, 1931), we find that corner pockets, though very small, may be fairly common and indicative of a crystal that experienced some sublimation before regrowth during its descent.
We also apply this protruding growth process to explain other features we observed, such as 'petal pockets', hollow terracing, and hollow-closing. In the case of hollow-closing, the area of the face also increases, though in the center region. Thus, the commonly observed center pockets are also a likely product of protruding growth.
We also suggest that the protruding growth mechanism may help us understand how some other snow-crystal types form, such as the sheath and scroll forms, as well as capped and multiple-capped columns. The mechanism may also have a significant influence on the maximum dimensions of thin, fast-growing forms such as the fast needles observed by Knight (2012) and fernlike dendrites.