Skip To Content
Cambridge University Science Magazine
The concept of metallic hydrogen has fascinated researchers ever since it was first predicted to exist almost a century ago, based on theoretical calculations. Perhaps this fascination is because of the puzzling idea that enough pressure could turn something so stubbornly gaseous as hydrogen into a metal. Perhaps it was the technical challenge of creating pressures only seen inside giant planets, under which even the diamond pressure vessels containing it would explode at the slightest provocation. Or perhaps it was the promise that metallic hydrogen could remain stable at more realistic pressures, enabling the possibility that superconducting cables of metallic hydrogen could be deployed to power the world more efficiently, as well as providing the vehicles of the future with a compact source of fuel.

Regardless of the many reasons driving interest in metallic hydrogen, the possibility for its creation has always been hampered by a lack of reproducibility. The difficulties of monitoring the properties of hydrogen under such extreme conditions without damaging its containment chamber meant that experimental teams often resorted to ad hoc means of doing so tailored around their unique setups. Naturally, this meant that very little research was available that could definitively pin down the range of temperatures and pressures under which metallic hydrogen could exist.

The authors sidestep these issues by resorting to computer modelling. However, computer modelling is very slow. Direct simulation of the properties of hydrogen over the required timescales would take an extremely long period of time, due to the fact that quantum effects must be taken into account. Add this to the fact that, in order to reconstruct a phase diagram, the authors need multiple simulations at each pressure. In addition, quantum systems become increasingly difficult to solve on non-quantum (normal) computers as the size of the system increases, given that the interactions of each atom with every other atom must be computed. The authors avoid this scalability issue by approximating the net effect of these interactions for each hydrogen atom given its position and velocity, taking into account general parameters such as temperature and pressure. To further speed things up, the authors use machine learning to extrapolate these approximations over a broader range of conditions.

The authors show that - as previously suspected - there is no definite point at which liquid hydrogen freezes or boils. Much like water, it can be heated or cooled well beyond its nominal boiling or freezing point while remaining a liquid. At higher pressures, the range of temperatures over which liquid hydrogen remains stable gradually narrows until only solid metallic and gaseous non-metallic hydrogen are stable. Hydrogen also remains metallic for much of the liquid state. Unexpectedly, liquid metallic hydrogen does not abruptly become ordinary, non-metallic liquid hydrogen when heated. Instead, the proportion of liquid hydrogen that is non-metallic gradually increases with temperature over the span of a few hundred degrees Centigrade until it eventually became entirely non-metallic.

Unfortunately for its potential applications, the presence of this gradual transition from a metallic phase into a non-metallic phase means that liquid metallic hydrogen cannot exist at pressures lower than that needed for it to form. In a gradual transition, the final phase coexists with the initial phase, while in an abrupt transition, the final phase must spontaneously form. In the latter case, unless conditions go too far beyond those needed for the stability of the initial phase, it needs some extrinsic factor to transition, such as an impurity or defect in the container. During the time this transition is waiting to happen, it is considered to be in a metastable state, where it will change state at the slightest provocation. If metallic hydrogen was to remain metastable down to reasonably achieved pressures, it would still be useful, but the gradual transition between metallic and non-metallic states precludes that from happening. Furthermore, while the authors do not address whether solid hydrogen exhibits a similar gradual transition from a metallic to a non-metallic state, previous calculations from first principles have ruled out its existence at easily attainable pressures. Specifically, below about ~10 GPa, which is the pressure you’d expect in the bottom of the Earth’s crust, solid metallic hydrogen becomes unstable. This is too high to be useful.

The presence of a gradual transition also forces us to reconsider estimates on the amounts of metallic hydrogen at different depths within gas giants. This is important as it constrains the sizes of the conductive metallic hydrogen mantles of gas giants, which are responsible for generating the magnetic fields that control how much of their atmospheres will be stripped off by stellar winds over time. More importantly, this research proves that neural networks are adequate to model quantum interactions; this potentially paves the way for their broader use in enabling more extensive quantum chemistry simulations, with benefits for the fields of condensed matter physics and biomedicine, to name but a few.

Research article: Cheng, B., Mazzola, G., Pickard, C.J. et al. Evidence for supercritical behaviour of high-pressure liquid hydrogen. Nature 585, 217–220 (2020).

Press release: AI shows how hydrogen becomes a metal inside giant planets

Clifford Sia is a medical student at the University of Cambridge. Photo of a diamond anvil cell by Steven Jacobsen at Northwestern University.