3.95-billion-year-old rock from China hints at very early subduction

Zircon crystals in a pale gneiss from eastern Hebei date to 3,951 ± 3 million years ago and carry chemical traces that, in the authors’ models, point to seafloor rock buried more than 50 km down and partly melted, which they read as a brief episode of subduction.

Peer reviewed · Measurements of rock samples · Subduction is an interpretation

A team led by Hangqiang Xie and Yusheng Wan of the Beijing SHRIMP Center at the Chinese Academy of Geological Sciences reports in Nature a 3.95-billion-year-old rock from eastern Hebei, North China. If the date holds, only Canada’s Acasta gneiss is older among rocks dated without dispute. Its zircon crystals point to two things: a mantle that had already been depleted by early crust-making in the Hadean, and seafloor crust buried deep enough to melt under high pressure. The catch: the subduction story rests on models of one small outcrop.

Evidence at a glance

Is this a big deal?

It is a solid step, not a settled answer. A new rock older than 3.9 billion years is rare news on its own, because only one other place, Acasta in Canada, has an undisputed one. The bigger claims about the mantle and subduction are careful inferences from a single outcrop. They deserve attention, and they now need a second site.

What scientists found

The rock is a trondhjemite gneiss: a pale, sodium-rich relative of granite, later squeezed and banded by heat and pressure. Rocks of this family are called TTGs (tonalite, trondhjemite and granodiorite), and according to the authors’ preprint most of the world’s 3.9-to-3.7-billion-year-old gneisses belong to it.

Its zircon crystals gave a weighted mean age of 3,951 ± 3 million years (95% confidence, 33 analyses). That date marks when the magma cooled. Younger zircon growth shows the rock was metamorphosed about 3.45 billion years ago, and today it survives in a block of ancient rock enclosed by 2.5-billion-year-old granite in the Labashan area of eastern Hebei.

One zircon analysis gave an even older 3,989 ± 10 million years. The authors do not treat it as the rock’s age: its lead pattern is slightly off, probably because later heating concentrated lead in that spot.

Two isotope signals carry the rest of the story. The zircons have a positive hafnium value, εHf(t) of +1.45 ± 0.27, the signature of magma whose ancestry runs back to a depleted mantle. And their oxygen is slightly heavy, with δ¹⁸O averaging +6.77 ± 0.23‰, above the roughly +5.3‰ typical of zircon from mantle-derived magma.

How they found it

The hills around Qian’an and Lulong were already known for old rock. A 2024 study led by Chunyan Dong, co-first author of the new paper, had reported granitoids aged 3.84 to 3.64 billion years there. This time they sampled a gneiss exposed on a hillside in the Shimen Complex: two fairly uniform samples and one banded sample threaded with thin, younger granite veins.

They dated the zircons with SHRIMP, an ion microprobe that fires a fine beam at a single crystal and counts the uranium and lead it releases. They then measured oxygen and hafnium isotopes on the same spots, along with trace elements in the crystals and the whole rock.

The oxygen works like a watermark. Basalt that soaks in cool seawater near the seafloor picks up a little extra oxygen-18, and magma made by melting that basalt keeps the mark. Like a watermark, though, it shows contact with water without saying exactly when or where it happened.

To estimate depth, Rongfeng Ge of Nanjing University and Xie modelled how basalt melts at different pressures and temperatures. After comparing early mafic rocks worldwide, they used an amphibolite from Acasta, which the authors note is the only known mafic rock older than 3.95 billion years. The Hebei chemistry matched melts made by melting 8–16% of that basalt at more than 1.6 GPa and 740–820 °C.

That pressure is about 16,000 times sea-level air pressure. Under rock with a density of about 3,100 kg/m³, it corresponds to roughly 53 km of overburden, in line with the more than 50 km given in Nature‘s Research Briefing.

Show the maths: from 1.6 GPa to kilometres

Pressure at depth is roughly density × gravity × depth. Rearranged: depth ≈ 1.6 × 10⁹ Pa ÷ (3,100 kg/m³ × 9.8 m/s²) ≈ 52,700 m, or about 53 km. The 3,100 kg/m³ is an assumed average for crust and upper mantle. A denser 3,300 kg/m³ gives about 49 km and a lighter 2,900 kg/m³ about 56 km, so the answer is roughly 50 km either way, and deeper because the model asks for more than 1.6 GPa. For scale, 1.6 × 10⁹ Pa ÷ 101,325 Pa (one atmosphere) ≈ 15,800.

Why it matters to you

Nothing about daily life changes. This is a “where did we come from” result. Continents are the ground under every city and farm, and geologists still argue about when and how the first pieces formed and when plates began sinking into the mantle. One rock cannot settle that, but rocks this old are so scarce that each one shifts the debate.

Why subduction, and what else it could be

Subduction is the process in which one tectonic plate sinks beneath another, dragging surface rock down into the mantle. The authors argue it best explains three findings together: rock altered by cool water near the surface (measured), melting at high pressure and a cool temperature gradient on the way down (both modelled). Their abstract stops at “most likely” and describes the event as transient, meaning brief, not the continuous plate motion of today.

According to the authors’ preprint, other routes struggle. Melting at the base of a thickened oceanic plateau would not explain the surface alteration. Sinking “drips” of dense crust need older continental crust nearby to sink through. Those are the authors’ arguments, not independent tests.

In the Hebei gneiss, the authors report geochemistry typical of high-pressure TTGs. They contrast that with the roughly 4.0-billion-year-old Idiwhaa unit at Acasta, for which earlier work inferred a shallow, low-pressure origin. If both readings are right, Earth made its earliest crust in more than one way.

What the evidence does not show

• No direct view of a plate. Subduction is inferred from chemistry and models; no structure from a 3.95-billion-year-old plate boundary survives at the site.
• One small outcrop. The preprint describes the 3.95-billion-year-old gneiss as exposed over about 15 square metres.
• Melt or mixture? Pressure estimates assume the rock records its original melt. Reviewer Martin Guitreau questioned this, citing Eu and Sr anomalies; the authors responded with tests of crystal accumulation and fractionation, which they report cannot reproduce the rock’s chemistry.
• Model-dependent timing. When the mantle was depleted depends on assumptions. The base case gives about 4.35 billion years ago; the authors’ most lenient case still requires depletion no later than about 4.2 billion years ago.
• A stand-in source rock. The melting model uses an Acasta amphibolite because so few mafic rocks this old survive anywhere.

How this fits what we already knew

The oldest well-dated rock remains the Acasta gneiss of Canada’s Northwest Territories, dated by Samuel Bowring and Ian Williams to 4.00–4.03 billion years. Older single zircon crystals exist, but the USGS notes that their source rocks have not been found. Some Nuvvuagittuq rocks in Quebec gave samarium–neodymium ages of about 4.16 to 4.20 billion years in a 2025 Science study, yet the authors of that paper note that the belt’s age has been debated, from 3.75 to 4.3 billion years.

Evidence for an early depleted mantle has mostly come from neodymium-142, a decay product of a short-lived isotope. Positive hafnium values in zircons this old have been scarce, which led some researchers to argue that large-scale mantle depletion started later. The Hebei zircons add a clear hafnium data point to the earlier side of that argument.

SciQuest covered a related window on the early mantle last month: crystals that hint the Archaean mantle was growing more oxidized. Our guide to Earth’s layers explains the crust and mantle this story moves between.

What the reviewers and other geologists said

SciQuest has not yet obtained comment from an independent geologist, and we found no news coverage quoting one. The closest on-record views come from the published peer-review file, in which two named reviewers signed their reports.

Elizabeth Bell, reviewing the submitted manuscript, wrote that the positive hafnium values of the oldest zircons are “Earth’s oldest yet identified clearly depleted mantle signature.” Her main request was more discussion of whether separate age and hafnium measurements could be mismatched, since an age that is too old would inflate the hafnium value.

Guitreau first recommended major revision. He asked whether the 3.95-billion-year zircons might be inherited from an older rock and whether the gneisses represent true melts. After revision he wrote that all his points had been addressed satisfactorily and recommended acceptance. He also pointed out that transient subduction and early horizontal plate motion had been proposed before, so “first evidence” claims need care.

How other headlines framed it

• Nature‘s Research Briefing led with Earth’s early history and the age, and stated the depth (more than 50 km) and a mantle reservoir dating back “as early as” 4.3 billion years.
• The paper’s title says subduction is “implied by” the rocks.
• The authors’ June 2026 preprint was titled more boldly: China’s oldest rocks “reveal” 3.95-billion-year subduction. Peer review softened the verb, and a reviewer also asked for “suggesting” in place of “indicating” in the abstract.

What happens next

The authors have posted their data tables in an open Zenodo repository, so other labs can rerun the ages and isotope calculations. The real test is a second rock: another high-pressure TTG older than 3.9 billion years, ideally on a different continent, would make the subduction reading much harder to dismiss. Reviewer Bell also suggested silicon isotopes as a way to pin down the cause of the heavy oxygen.

In one sentence

A 3.95-billion-year-old gneiss from Hebei, the second-oldest well-dated rock on Earth, carries chemical marks of seafloor crust buried more than 50 km deep and melted, and its finders argue that only a brief bout of subduction could have put it there.

Quick check: did it click?

1. Which part of the study is a direct measurement: the age or the subduction? The age. Subduction is the authors’ interpretation of the chemistry and melting models.
2. What does the heavy oxygen in the zircons suggest? The basalt that melted had been altered by cool water near Earth’s surface before it was buried.
3. Is this the oldest rock on Earth? No. The Acasta gneiss in Canada, at 4.00–4.03 billion years, is older.

Read next

• Crystals hint Earth’s mantle was oxidizing before the Great Oxidation Event: another zircon record of the early mantle.
• Teghaza 001 meteorite samples Mars’s early crust: how another planet built its first crust.
• The Grand Canyon’s missing billion years: what happens to very old rock records.

Sources and disclosure

1. Xie H, Dong C, Li P, Ge R, Wilde SA, Liu S, Liu D, Wan Y. “Hadean mantle depletion and 3.95-Gyr subduction implied by North China rocks.” Nature, published 11 November 2026 (received 13 April, accepted 8 October 2026). doi:10.1038/s41586-026-11280-2. Abstract, figure and Extended Data captions read on the article page.
2. Peer Review File for the paper (CC BY 4.0), source of the Bell and Guitreau comments.
3. “Earth’s early history told by 3.9-billion-year-old rocks from northern China.” Nature Research Briefing, 11 November 2026. doi:10.1038/d41586-026-03499-w.
4. Xie H et al. “China’s oldest rocks reveal 3.95 Ga subduction and Hadean mantle depletion.” Research Square preprint, posted 8 June 2026, not peer reviewed. doi:10.21203/rs.3.rs-9401592/v1 (source of the outcrop size and the alternative-mechanism arguments).
5. Bowring SA, Williams IS. “Priscoan (4.00–4.03 Ga) orthogneisses from northwestern Canada.” Contributions to Mineralogy and Petrology 134, 3–16 (1999).
6. Sole C et al. “Evidence for Hadean mafic intrusions in the Nuvvuagittuq Greenstone Belt, Canada.” Science 388, 1431–1435 (2025). doi:10.1126/science.ads8461.
7. Valley JW et al. “4.4 billion years of crustal maturation: oxygen isotope ratios of magmatic zircon.” Contributions to Mineralogy and Petrology 150, 561–580 (2005) (mantle zircon δ¹⁸O of 5.3 ± 0.6‰).
8. Dong C et al. “New discovery of 3.84–3.64 Ga diverse granitoids in eastern Hebei, North China Craton.” GSA Bulletin 136, 5249–5261 (2024).
9. US Geological Survey. Age of the Earth (4.54 billion years).

Disclosure: This article was drafted with AI assistance and checked against the sources above; the paper’s main text is paywalled, so details come from its abstract, figure captions, Extended Data, peer-review file and the authors’ preprint, and preprint-only details are labelled. SciQuest calculations: 1.6 GPa ÷ 101,325 Pa ≈ 15,800 times sea-level pressure; depth ≈ 1.6 GPa ÷ (3,100 kg/m³ × 9.8 m/s²) ≈ 53 km, assuming an average rock density; the one-day clock scales 4.54 billion years to 24 hours. The graphic is a SciQuest original built from these sources. Spotted an error? See our corrections policy.