Cratonic lithosphere is believed to have been chemically buoyant and mechanically resistant to destruction over billions of years. Yet the absence of cratonic roots at some Archean terrains casts doubt on the craton stability and longevity on a global scale. As unique mantle-derived melts at ancient continents, silica-poor, kimberlitic melts are ideal tools to constrain the temporal variation of lithosphere thickness and the processes affecting the lithosphere root. However, no reliable thermobarometer exists to date for strongly silica-undersaturated, mantle-derived melts. Here we develop a new thermobarometer for silica-poor, CO₂-rich melts using high-temperature, high-pressure experimental data. Our barometer is calibrated based on a new observation of pressure-dependent variation of Al₂O₃ in partial melts saturated with garnet and olivine, while our thermometer is calibrated based on the well-known olivine-melt Mg-exchange. For applications to natural magmas, we also establish a correction scheme to estimate their primary melt compositions.

Applying this liquid-based thermobarometer to the estimated primary melt compositions for a global kimberlite dataset, we show that the equilibration depths between primary kimberlite melts and mantle peridotites indicate a decrease of up to ∼150 km in cratonic lithosphere thickness globally during the past ∼2 Gyr. Together with the temporal coupling between global kimberlite frequency and cold subduction flux since ∼2 Gyr ago, our results imply a causal link between lithosphere thinning and supply of CO₂-rich melts enhanced by deep subduction of carbonated oceanic crusts. While hibernating at the lithosphere root, these melts chemically metasomatize and rheologically weaken the rigid lithosphere and consequently facilitate destruction through convective removal in the ambient mantle or thermo-magmatic erosion during mantle plume activities.