Ultrahigh-pressure metamorphism is the process by which crustal rocks, the buoyant material of continents, are dragged down to depths greater than 90 kilometres, deep enough for the mineral coesite, a dense high-pressure form of quartz, to become stable. Because reaching those depths typically requires rapid, efficient transport into the mantle, geologists have long treated the presence of coesite as evidence for cold subduction, where the descending slab stays cooler than its surroundings. Peak metamorphic conditions recorded by such rocks usually imply low temperature-to-pressure gradients of less than 330 degrees Celsius per gigapascal, consistent with thermal models of mature subduction zones. Yet, as the new study points out, direct mineralogical evidence for a cold prograde journey, the actual path a rock takes on its way down, has been conspicuously lacking. Overprinting during burial and exhumation tends to erase the early record, so the cold prograde gradient has always been an assumption inferred from peak conditions rather than an observation.

The rock at the centre of the story, sample 03118b, was collected from eastern Taleba Bay on southern Goodenough Island, close to a beach placer where coesite-bearing detrital garnets had previously been discovered. It forms a small lens of mafic composition, roughly 90 by 30 centimetres, hosted within felsic gneiss. The eclogite is partially retrogressed, meaning it was partially altered as it returned toward the surface, but its garnet crystals remain pristine archives of the burial history. Garnet is a remarkable mineral for this purpose because it grows in concentric zones, each layer locking in whatever minerals, melts and fluids were present at the moment of crystallization. In this sample, the garnet displays three distinct compositional zones: a calcium-rich core, a mantle zone with lower calcium, and a rim with rising calcium again, each zone corresponding to a different stage of the subduction journey.

The inclusion inventory within the garnet zones provides the crucial mineralogical evidence. The cores contain quartz, rutile, zircon, apatite, omphacite and titanite, but no coesite. The mantle zones, by contrast, host monomineralic coesite inclusions measuring roughly one to four micrometres, along with slightly larger bimineralic coesite-quartz inclusions of about five micrometres. Because coesite is only stable above the quartz-to-coesite transition at approximately 2.7 to 3.0 gigapascals, the boundary between core and mantle marks the moment the rock crossed into the ultrahigh-pressure stability field. The size difference between the two inclusion types also supports the idea that larger inclusions preserve coesite better, since the surrounding garnet exerts protective overpressure on its trapped cargo.

Even more revealing are the polyphase inclusions found in both core and mantle. These tiny droplets, termed nanogranitoids, consist of silica polymorphs such as quartz or cristobalite, feldspar polymorphs including kumdykolite, kokchetavite and dmisteinbergite, and phyllosilicates such as muscovite, phlogopite and chlorite, sometimes accompanied by magnetite and carbonaceous matter. Their peculiar metastable mineralogy is the signature of a silicate melt that was trapped as a liquid and then crystallized in place within the confined space of the inclusion. In other words, the garnet grew in the presence of melt, meaning the rock was above its solidus, the temperature at which partial melting begins, throughout its burial. That single observation carries enormous thermal implications, because melting of basic rocks requires temperatures exceeding roughly 670 degrees Celsius.

To pin down the exact conditions, the team deployed an impressive analytical arsenal. Raman spectroscopy mapped the inclusions in two and three dimensions at sub-micrometre resolution. Zirconium-in-rutile thermometry on rutile inclusions yielded temperatures from the Zr content of the tiny crystals. Most dramatically, the researchers performed re-homogenization experiments using a piston-cylinder apparatus at GFZ Potsdam: garnet chips containing crack-free nanogranitoids were compressed to 2.5 gigapascals and heated to 750 degrees Celsius for 24 hours, successfully melting the crystallized inclusions back to glass without any reaction with the host garnet. This experiment demonstrated that the garnet core grew at approximately 700 to 750 degrees Celsius. Combined with thermodynamic modelling of the garnet composition, the team arrived at a best estimate for core formation of 1.5 plus or minus 0.2 gigapascals and 700 plus or minus 20 degrees Celsius, while the mantle zone, constrained by coesite and rutile thermometry, formed at pressures greater than 2.9 gigapascals and about 800 plus or minus 25 degrees Celsius.

Dating the journey required extracting micrometre-sized zircon inclusions from the garnet, a feat accomplished by dissolving the host mineral and handpicking the roughly ten-micrometre zircons under a stereo microscope. Using a large-geometry secondary ion mass spectrometer at Curtin University, the team analysed 26 grains, of which 17 yielded reliable uranium-lead ages spanning 5.1 to 3.1 million years. The trace-element chemistry of the zircons allowed the dates to be tied to specific garnet growth zones. All grains show low thorium-to-uranium ratios below 0.05, no negative europium anomaly and flat heavy rare-earth-element patterns, indicating growth in equilibrium with garnet, which preferentially partitions heavy rare earths. From 5.1 to 4.4 million years ago, heavy rare-earth concentrations declined as garnet grew in abundance, marking the interval of subduction from pressures of at most 1.5 gigapascals to greater than 2.9 gigapascals.

That timing yields a staggering result. Covering the pressure interval from 1.5 to 2.9 gigapascals, equivalent to roughly 45 to more than 90 kilometres of depth, in only about 700,000 years implies vertical subduction rates exceeding 60 millimetres per year, assuming lithostatic pressure. The ultrahigh-pressure stage itself, dated at about 4.4 million years, is considerably younger than the roughly 7.1-million-year coesite eclogite previously reported from Tumagabuna Island nearby, confirming that the Woodlark Rift region hosts Earth’s youngest known coesite eclogites. The rapid burial also explains why the earlier prograde record survived at all: the rock was whisked down and back up before diffusion and overprinting could erase the evidence.

When the pressure-temperature data are assembled into a path, the thermal transition becomes unmistakable. The shallow prograde path, from the surface to about 1.5 gigapascals, follows an apparent gradient of roughly 470 degrees Celsius per gigapascal, far warmer than any mature subduction zone model predicts, even for the warmest modern examples such as Mexico and Cascadia. Then, beyond about 45 kilometres of depth, the path pivots sharply into a cold regime consistent with gradients below 330 degrees Celsius per gigapascal, matching the thermal structure of mature subduction zones. The team considered whether the core and mantle might record two separate metamorphic events, but the tight zircon age range of 5.1 to 3.1 million years supports a single continuous burial episode. Intriguingly, published pressure-temperature paths from other ultrahigh-pressure terranes such as the Sanbagawa belt of Japan and the Dabie-Sulu orogen of China show similar warm-then-cold shapes, suggesting this may be a common feature rather than a local curiosity.

Two mechanisms, not mutually exclusive, can explain the warm start. The first is shear heating along the plate interface: as the descending plate grinds against the overriding plate, frictional energy dissipates as heat, warming the interface rocks. Thermal models that include shear heating with an effective coefficient of friction of about 0.07 produce interface temperatures closer to the observed path, though still somewhat cooler, and the present-day convergence between the Pacific and Australian plates at roughly 110 millimetres per year could push temperatures higher still. The second mechanism is the inherent time-dependence of young subduction zones. Numerical models show that in the first several million years after subduction initiates, before a quasi-steady-state thermal structure establishes, rocks entering the trench follow markedly warmer and more variable paths. The observed path closely resembles model predictions for rocks entering a subduction zone about 5 to 12 million years after initiation, and the inferred subduction rate of more than 60 millimetres per year matches models for rocks entering about 8 to 13 million years after initiation. The authors caution that peak metamorphic conditions alone, which in this rock imply a gradient of only about 280 degrees Celsius per gigapascal, can be profoundly misleading: some high-temperature rocks may have formed in zones that later turned cold, and some low-gradient rocks may never have experienced a cold, steady-state subduction system at all. For geologists using metamorphic rocks to reconstruct Earth’s tectonic past, the garnets of Goodenough Island deliver a clear message: the journey matters as much as the destination.

Subject of Research: Geothermal gradient evolution during early subduction recorded by a young ultrahigh-pressure eclogite from Papua New Guinea

Article Title: Geothermal gradient change during subduction recorded by ultrahigh-pressure eclogite

Article References: Schönig, J., Ferrero, S., Gresky, K., Sorger, D., Schmitt, A. K., Wunder, B., Baldwin, S. L., & Holt, A. F. (2026). Geothermal gradient change during subduction recorded by ultrahigh-pressure eclogite. Nature Geoscience. https://doi.org/10.1038/s41561-026-02110-1

Image Credits: AI Generated

DOI: 10.1038/s41561-026-02110-1

Keywords: subduction zones, ultrahigh-pressure metamorphism, eclogite, coesite, garnet, geothermal gradient, zircon U-Pb dating, melt inclusions, shear heating, Papua New Guinea, thermobarometry, plate tectonics