MAGMA FLOWS EASIER WHEN SQUEEZED — BUT THAT'S NOT WHY IT POOLS

Earth’s tectonic plates slide over the asthenosphere, a layer of the upper mantle that is unusually soft. One long-standing idea is that small amounts of molten rock — basaltic magma — help soften it. How easily that melt moves at depth depends on its viscosity. And there, a strange result has divided geoscientists.

Compress most materials and they become stiffer. Yet high-pressure experiments, notably by Sakamaki and colleagues in 2013, reported that molten basalt becomes less viscous under pressure, down to a minimum at depths of roughly 90 to 150 kilometres, before stiffening again. The idea has been used to explain how magma moves, seismic oddities in the mantle and layered chemistry. In 2025, Russell and colleagues countered that the minimum was an experimental artefact, caused by poorly controlled temperatures: corrected, they argued, viscosity simply rises with pressure.

Why experiments struggle

At these pressures, temperature and pressure are hard to separate. Thermometers in high-pressure presses carry uncertainties of 50 to 100 kelvin or more, and the classic method — timing a tiny ball falling through the melt — is disturbed by the walls of millimetre-sized capsules. The effect being sought is subtle; the noise is not.

Billions of atomic steps

Hongkun Zeng, Liang Yuan and colleagues at the China University of Geosciences in Wuhan, the University of Bayreuth and Sichuan University turned to simulation. They modelled a simplified basalt (a mix of the minerals diopside and anorthite) atom by atom, with forces computed from quantum mechanics and imitated by a neural network, which lets the simulation run more than 1,000 times longer than standard quantum methods.

The key was statistics. For each pressure and temperature they ran 150 to 600 independent simulations of 5 nanoseconds each, totalling at least 7.5 microseconds per path, after checking that their method reproduces the measured densities of real minerals to within 1%. That brought the uncertainty on viscosity down to 1 to 3%, against up to 50% in earlier studies.

A shallow U

At a constant 2,073 kelvin, the viscosity traces a gentle U: 0.108 pascal-seconds at zero pressure, a minimum of 0.088 at 3 gigapascals, then 0.104 at 6 gigapascals — a dip of about 19%.

Along the melting curve, as in experiments, viscosity instead falls by a factor of about 23 between 0 and 6 gigapascals — but that is mainly the effect of rising temperature. At zero pressure, cooling from 2,073 to 1,619 kelvin multiplies the viscosity by about 18. A typical experimental error of 100 kelvin thus shifts viscosity by more than twenty times the size of the U. No wonder it was hard to see.

The cause lies in the atomic structure. Silicon stays mostly locked in rigid groups of four oxygens. Aluminium, by contrast, rearranges strongly under pressure — its groups of five oxygens rise from about 20% to 50% of the total — and these rare, fast-changing aluminium sites match the time it takes the melt to relax. They, not the silicon framework, set the pace of the flow.

Heat, not the dip

The team then modelled how melt percolates upward through the mantle. Its mobility rises with depth, peaks near 150 kilometres, and falls as the melt climbs and cools, so magma tends to stall and accumulate just below the rigid plate — consistent with seismic and electrical signs of a melt-rich layer at 45 to 70 kilometres depth.

But the authors stress a nuance their own abstract downplays: this mobility peak comes from the competition between pressure and temperature along the Earth’s temperature profile, not from the 3-gigapascal minimum. The viscosity minimum is real, they conclude, but “neither necessary nor sufficient” to explain the ponding of melt; temperature alone does the job. Their basalt is simplified — no iron, no sodium, water as the only volatile — and the minimum’s position depends on composition.