The division of the earth's surface into continents and oceans is a consequence of plate tectonics but a geological paradox exists at continent-ocean boundaries. Continental plate is thicker and lighter than oceanic plate, floating higher on the mantle asthenosphere, but it can rift apart by thinning and heating to form new oceans. In theory, continental plate subsides in proportion to the amount it is thinned and subsequently by the rate it cools down. However, seismic and borehole data from continental margins like the Atlantic show that the upper surface of many plates remains close to sea-level during rifting, inconsistent with its thickness, and subsides after breakup more rapidly than cooling predicts. Here we use numerical models to investigate the origin and nature of this puzzling behaviour with data from the Kwanza Basin, offshore Angola. We explore an idea where the continental plate is made increasingly buoyant during rifting by melt produced and trapped in the asthenosphere. Using finite element simulation, we demonstrate that partially molten asthenosphere combined with other mantle processes can counteract the subsidence effect of thinning plate, keeping it elevated by 2-3 km until breakup. Rapid subsidence occurs after breakup when melt is lost to the embryonic ocean ridge. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ . Abstract The division of the earth’s surface into continents and oceans is a consequence of plate tectonics but a geological paradox exists at continent-ocean boundaries. Continental plate is thicker and lighter than oceanic plate, floating higher on the mantle asthenosphere, but it can rift apart by thinning and heating to form new oceans. Using finite element simulation, we demonstrate that partially molten asthenosphere combined The established rift paradigm explains the formation of ocean basins by stretching, thinning and breakup of continental plate or lithosphere. As the crust thins, it subsides due to decreasing buoyancy while the asthenosphere rises to fill the extensional space created beneath the necking lithosphere. When rifting ceases, the crust then sinks further as it and the underlying mantle cool, a model elegantly enumerated by McKenzie 1 . Furthermore, we do not know of any models designed specifically to solve the issue of rapid sag after breakup, although explanations have been proposed such as continued extension below the crust 7 , 13 , 14 . We address this shortfall here, by investigating mechanisms which might keep the plate buoyant as it thins and subsequently cause it to subside quickly. Compared to the crust, the mantle is thick and therefore has the potential to significantly influence the buoyancy of the plate. One plausible scenario explaining the anomalous subsidence patterns is that the mantle gets lighter as rifting proceeds 15 , 16 . Although it becomes thinner as it stretches, the plate in continental rifts is composed of granitic crust underlain by thick mantle lithosphere. This mantle lithosphere neither melts nor circulates easily and is therefore chemically and rheologically different to the mantle asthenosphere beneath the plate. It is also colder than the asthenosphere, conducting rather than convecting heat and is relatively impermeable to melt 25 , magma transported upwards along fractures and faults rather than by melt flowing through a pore network 19 , 20 , 23 . In contrast, along the rift axis of most mid-ocean ridges the mantle lithosphere is absent: the ridge connected instead to shallow, upwelling asthenosphere where the melt responsible for oceanic crust is generated 26 , which flows upwards through pores 11 , 27 . We now hypothesize that the relative impermeability of mantle lithosphere to melt may be the root cause of additional elevation. Thus, as continental plate thins during rifting, the asthenosphere wells up and pressure is reduced causing it to partially melt in proportion to the height the asthenosphere rises 17 , 25 , 26 . Melt is lower density than the parent asthenosphere 28 and will tend to migrate upwards through the asthenosphere by compaction once a percolation threshold defined by melt-filled porosity is crossed 29 . However, the overlying mantle lithosphere will act as a barrier, impeding its upward flow 12 , 19 , 23 , 25 . This opens the possibility that significant amounts of melt are pooled in the asthenosphere 30 , 31 making it buoyant 17 so that the overlying lithosphere rises 18 , 19 , 22 , 23 . When the rifting plate breaks, the melt is then lost to the embryonic ocean ridge as seafloor spreading starts. Also, by themselves, no other effects are sufficient to stop the surface of the plate sinking to oceanic depths at the point of breakup (Fig. 1a ). It is worth noting that the thicknesses of underplates produced in models Fdd, Fdd15o and Fdd130 (Fig. 1b ) are similar to those of high velocity, high density bodies seen at the base of many rifted basins 41 , including the South Atlantic 42 , 43 . They are generally reflective with a layered appearance, atypical of serpentinized mantle 44 and are generally interpreted as underplates or layers of mafic and ultramafic intrusions although ref. 45 leaves the question open. Nonetheless, we test the draw-down idea with two models UFB (underfilled basin) and BL-250 (base-level 250 m below global sea-level) and others in Supplementary Information but none of these come close to achieving a good stratigraphic match or the correct depositional water depth conditions (Table 1 and Fig. 1a ). Therefore, as well as testing different tectonic and stratigraphic scenarios, the type of basin modeling reported here can be used to help verify or reject plate reconstructions.