Résumé : A thin glacial diamicton, informally termed Granite drift, occupies the floor of central Beacon Valley in southern Victoria Land, Antarctica. This drift is <1.0 m thick and rests with sharp planar contacts on stagnant glacier ice reportedly of Miocene age, older than 8.1 Ma. The age of the ice is based on 40Ar/39Ar analyses of presumed in situ ash-fall deposits that occur within Granite drift. At odds with the great age of this ice are high-centered polygons that cut Granite drift. If polygon development has reworked and retransported ash-fall deposits, then they are untenable as chronostratigraphic markers and cannot be used to place a minimum age on the underlying glacier ice.Our results show that the surface of Granite drift is stable at polygon centers and that enclosed ash-fall deposits can be used to define the age of underlying glacier ice. In our model for patterned-ground development, active regions lie only above polygon troughs, where enhanced sublimation of underlying ice outlines high-centered polygons. The rate of sublimation is influenced by the development of porous gravel-and-cobble lag deposits that form above thermal-contraction cracks in the underlying ice. A negative feedback associated with the development of secondary-ice lenses at the base of polygon troughs prevents runaway ice loss. Secondary-ice lenses contrast markedly with glacial ice by lying on a δD versus δ18O slope of 5 rather than a precipitation slope of 8 and by possessing a strongly negative deuterium excess. The latter indicates that secondary-ice lenses likely formed by melting, downward percolation, and subsequent refreezing of snow trapped preferentially in deep polygon troughs.The internal stratigraphy of Granite drift is related to the formation of surface polygons and surrounding troughs. The drift is composed of two facies: A nonweathered, matrix-supported diamicton that contains >25% striated clasts in the >16 mm fraction and a weathered, clast-supported diamicton with varnished and wind-faceted gravels and cobbles. The weathered facies is a coarse-grained lag of Granite drift that occurs at the base of polygon troughs and in lenses within the nonweathered facies. The concentration of cosmogenic 3He in dolerite cobbles from two profiles through the nonweathered drift facies exhibits steadily decreasing values and shows the drift to have formed by sublimation of underlying ice. These profile patterns and the 3He surface-exposure ages of 1.18 ± 0.08 Ma and 0.18 ± 0.01 Ma atop these profiles indicate that churning of clasts by cryoturbation has not occurred at these sites in at least the past 105 and 106 yr. Although Granite drift is stable at polygon centers, low-frequency slump events occur at the margin of active polygons. Slumping, together with weathering of surface clasts, creates the large range of cosmogenic-nuclide surface-exposure ages observed for Granite drift. Maximum rates of sublimation near active thermal-contraction cracks, calculated by using the two 3He depth profiles, range from 5 m/m.y. to 90 m/m.y. Sublimation rates are likely highest immediately following major slump events and decrease thereafter to values well below our maximum estimates. Nevertheless, these rates are orders of magnitude lower than those computed on theoretical grounds. During eruptions of the nearby McMurdo Group volcanic centers, ash-fall debris collects at the surface of Granite drift, either in open thermal-contraction cracks or in deep troughs that lie above contraction cracks; these deposits subsequently lower passively as the underlying glacier ice sublimes. The fact that some regions of Granite drift have escaped modification by patterned ground for at least 8.1 Ma indicates long-term geomorphic stability of individual polygons. Once established, polygon toughs likely persist for as long as 105–106 yr. Our model of patterned-ground formation, which applies to the hyperarid, cold-desert, polar climate of Antarctica, may also apply to similar-sized polygons on Mars that occur over buried ice in Utopia Planitia.