Q: How does hydrogen turn into a metal? A: Hang on a second, I need to train my AI supercomputer first

Since we can't build a planet

Scientists have trained a neural network on a supercomputer to simulate how hydrogen turns into a metal, an experiment impossible to reproduce physically on Earth.

Under extreme pressures and high enough temperatures – such as in the cores of Jupiter, Saturn, Uranus, and Neptune – hydrogen enters a strange phase. The electrons normally bound to its nuclei are free to move, and they collectively whiz around to conduct electricity, a common property in metals.

The physics behind the process is difficult to study. Attempting to replicate the conditions inside those planet cores here on Earth is pointless – the sheer amount of energy required is impractical. Instead, the best way to study the phase transitions is to recreate it using software.

Modelling the interactions between each hydrogen atom, however, is no easy task, and even the most powerful supercomputers struggle. “Both electrons and nuclei follow the laws of quantum mechanics, so their behavior can be described by solving the Schrodinger equation,” Bingqing Cheng, the lead author of the study just published in Nature and a research fellow at the University of Cambridge explained to The Register. A preprint version is here, and source code here.

Supercomputer, image via shutterstock

Microsoft claims it has spun up a top-five AI supercomputer for its pals at OpenAI – but won't reveal the full specs


“As the number of electrons and nuclei increases, the complexity involved soon becomes intractable even with the fastest supercomputers. In fact, quantum mechanical calculations are still unaffordable for systems with more than a few hundred atoms.”

Here’s where the machine-learning algorithms come in handy: they can be trained to predict the states of hydrogen atoms so that supercomputers don’t have to simulate all the interactions between the particles.

Without the neural network, it would take "several hundred millions of CPU years" to carry out the same simulations, according to the paper. “We exploited an artificial neural network (ANN) to learn the atomic interactions from quantum mechanics. The ANN first learns quantum mechanical interactions between atoms, and then makes speedy predictions about the energy and forces for a system of atoms, bypassing the need to perform expensive quantum mechanical calculations,” Cheng told us on Wednesday.

Cheng and her colleagues at IBM Research and the Swiss Federal Institutes of Technology in Lausanne, Switzerland, discovered that hydrogen undergoes a smooth transition when it becomes a metal. In other words, the change is gradual depending on its pressure and temperature. The hydrogen inside the aforementioned giant planets can exist in multiple states or layers; some regions might be metallic and others might not be.

At increasing temperatures and pressures, more and more hydrogen turns into a metal. “Roughly speaking, [this] pressure is about 200 GPa at 1000K, and decreases a bit as temperature goes up,” Cheng said.

Unlocking the secrets of how hydrogen morphs into an electricity-conducting metal could help scientists create room temperature superconductors one day or understand the magnetic fields of the outer planets in the Solar System.

“The metallic nature of hydrogen at high pressure is particularly important, as it is the reason why both Jupiter and Saturn have strong magnetic fields. Metallic hydrogen [also] has a number of exotic properties, such as superconductivity at room temperature and superfluidity. Understanding this material is thus useful for potentially utilizing this super material,” Cheng concluded. ®

Similar topics


Send us news

Other stories you might like