AI simulations draw a colder boundary for helium rain inside giant planets

Deep inside a giant planet, helium may gather into droplets and sink through hydrogen. The descent can release gravitational energy as heat and leave the layers above depleted in helium. This proposed process helps connect an otherwise invisible interior to a planet’s warmth and atmospheric composition.[3]

Calculations published on 1 September move that boundary typically about 2,000 kelvin below earlier small-system estimates. Xiaoyu Wang, Sebastien Hamel and Bingqing Cheng infer that separation is plausible in Saturn but unlikely in Jupiter under the interior profiles examined.[1]

Rain without a weather cycle

The word “rain” describes precipitation and settling, not clouds recycling through an atmosphere. Nor are the droplets necessarily pure helium: separation produces a helium-rich fluid and a helium-poor one. Their proportions depend on how much helium the original mixture contains. Pressure, temperature and composition together determine whether splitting is favourable.[5]

Think of the question as whether one uniform mixture is more stable than two fluids of different composition. Temperature encourages mixing by increasing the importance of the many ways atoms can arrange themselves. Atomic interactions can favour separation. A phase diagram marks where that balance changes. Knowing the ingredients alone is therefore insufficient to predict what a planet’s interior will do.

The new calculation gives 3,511 K at 150 gigapascals for a helium atomic fraction of 0.089, using the PBE approximation to quantum interactions. That fraction counts atoms, rather than helium’s share of mass.[1]

Giving the atoms enough room

Calculating quantum interactions afresh as atoms move is expensive. The researchers train machine-learning potentials to reproduce those interactions, then use the learned rules to follow larger systems. Their main production simulations contain 3,456 atoms and run for 60 picoseconds, across a pressure grid of 100–1,000 GPa. A picosecond is a trillionth of a second: this is an atomic motion calculation, not a simulation of a planet’s entire cooling history.[2]

Size matters because a mixture must have room to develop variations in concentration. In tests of this method, key thermodynamic results stabilised at around 1,728 atoms and above. Smaller boxes could suppress the fluctuations used to infer separation. The team also checked its learned interactions against fresh quantum calculations of the mixing free energy, the quantity that decides which arrangement is favoured.[2]

The AI’s contribution is consequently quite specific. It makes a larger physical calculation affordable; the scientific interpretation comes from asking which combinations of temperature, pressure and composition can coexist. A vivid animation of two colours separating would not, by itself, supply that boundary.

Two planets take different paths

The Jupiter profiles examined largely stay too warm or too helium-poor for separation. Cooler Saturn profiles can enter the demixing region, particularly where helium accumulates. The inference depends on those interior profiles as well as the atomic calculation.[1]

Once separation begins, it also changes the conditions for further separation. Sinking material removes helium from one region and enriches another; heat transport and the planet’s cooling history can change with it. A single mixture calculation supplies an ingredient for that evolving planetary model. It does not specify a rainfall rate or tell us exactly when the process began.[5]

The disagreement is part of the discovery

An earlier machine-learning study, published in 2024 by Xiaoju Chang and colleagues, reached a hotter boundary and inferred demixing in both planets. Its relevant production simulations used 27,000 atoms. It classified separation through statistics of which atoms occupied one another’s neighbourhoods, and performed additional free-energy checks. The dispute therefore involves how separation is measured and interpreted, not a simple contest between large and small simulations.[3]

The newer authors suggest that finite-size assumptions and the resulting free energies may explain discrepancies with older small-cell calculations. In their peer-review responses, they explicitly describe this as a hypothesis that would require fuller replication of the earlier work to establish. It is not a demonstrated explanation for every competing result.[5]

A 2021 Nature experiment pre-compressed hydrogen–helium in diamond-anvil cells, then shock-compressed it with a laser. Reflectivity changes indicated an immiscibility region ending above 150 GPa at 10,200 K. Its helium atomic fraction was 0.11. That composition and compression path differ from the 0.089, fixed-pressure calculation above, so these temperatures are not a matched comparison. The much hotter experimental boundary remains an important disagreement.[4]

The immediate puzzle is Jupiter: if its interior has remained too warm for helium to rain out, what accounts for the helium depletion measured in its atmosphere?[3]

Report an error or suggest a correction