Evidence for supercritical behavior of high-pressure liquid hydrogen

Bingqing Cheng, Guglielmo Mazzola, Michele Ceriotti

Abstract: Hydrogen exhibits unusual behaviors at megabar pressures, with consequences for planetary science, condensed matter physics and materials science. Experiments at such extreme conditions are challenging, often resulting in hard-to-interpret and controversial observations. We present a theoretical study of the phase diagram of dense hydrogen, using machine learning to overcome time and length scale limitations while describing accurately interatomic forces. We reproduce the re-entrant melting behavior and the polymorphism of the solid phase. In simulations based on the machine learning potential we find evidence for continuous metallization in the liquid, as a first-order liquid-liquid transition is pre-empted by freezing. This suggests a smooth transition between insulating and metallic layers in giant gas planets, and reconciles existing discrepancies between experiments as a manifestation of supercritical behavior.

Introduction

Hydrogen, the most abundant element in the universe, undergoes a dramatic transformation at the extreme pressures found within the interiors of giant planets. At megabar pressures, it transitions from a molecular insulator to an atomic metal—a process known as the Liquid-Liquid Transition (LLT). For decades, the nature of this transition has been a subject of intense debate: is it a sharp, first-order phase change or a smooth, continuous crossover? Experimental results have been notoriously difficult to reconcile, with reported transition pressures varying by as much as 100 GPa.

Machine Learning for Quantum Accuracy

To resolve this controversy, researchers employed a Machine-Learning Potential (MLP) based on the Behler-Parrinello neural network framework. This approach allows for simulations that maintain the accuracy of Density Functional Theory (DFT) while being computationally efficient enough to model thousands of atoms over nanosecond timescales. The MLP was trained on a diverse dataset of over 9,000 hydrogen configurations and validated against a massive set of 200,000 structures, ensuring robust performance across a wide range of temperatures (100–4000 K) and pressures (25–400 GPa).

The Supercritical Crossover

The study's primary finding is that the liquid-liquid transition in hydrogen is supercritical. While a first-order transition exists at lower temperatures, it is "pre-empted" by the freezing of liquid hydrogen into a solid. Consequently, in the stable liquid phase above the melting line, the transition from molecular to atomic hydrogen is a smooth crossover.

Figure 3: Phase diagram of dense hydrogen (a) showing the supercritical region and Widom lines, alongside temperature-dependent density and heat capacity (b) and structural snapshots (c).

As shown in Figure 3, different physical properties—such as density ($\rho$) and heat capacity ($C_p$)—reach their maximum rates of change at slightly different pressures and temperatures. these loci of maxima are known as Widom lines. This phenomenon explains the long-standing experimental discrepancies: different teams, measuring different properties, were essentially observing different Widom lines within the same supercritical region.

Thermodynamic Modeling and the Critical Point

To precisely locate the liquid-liquid critical point, the researchers utilized a polyamorphic solution model. By analyzing the Gibbs free energy ($g$) as a function of molecular fraction ($x$), they could distinguish between first-order behavior (characterized by a double-well potential) and supercritical behavior (a single-well potential).

Figure 6: Thermodynamic modeling of the transition, showing the evolution of Gibbs free energy curves (a) and the temperature dependence of model parameters (b) used to identify the critical point.

The analysis, illustrated in Figure 6, predicts the critical point to be at approximately 350 GPa and 380 K. Because this temperature lies below the melting point at that pressure, the liquid remains supercritical throughout its stable range. Additionally, the study successfully reproduced the "re-entrant" melting behavior of hydrogen and identified a novel solid phase at high pressures consisting of atomic "wires" in a hexagonal lattice.

Implications for Planetary Science

These findings have significant implications for our understanding of gas giants like Jupiter and Saturn. Traditional models often assume a sharp boundary between an outer insulating mantle and an inner metallic core. This research suggests that these layers likely transition into one another through a smooth gradient. Furthermore, the machine-learning framework developed here provides a powerful tool for future studies of hydrogen-helium mixtures, potentially addressing unresolved questions regarding the cooling history and luminosity of planets like Saturn.

Summary updated

Abstract

Hydrogen exhibits unusual behaviors at megabar pressures, with consequences for planetary science, condensed matter physics and materials science. Experiments at such extreme conditions are challenging, often resulting in hard-to-interpret and controversial observations. We present a theoretical study of the phase diagram of dense hydrogen, using machine learning to overcome time and length scale limitations while describing accurately interatomic forces. We reproduce the re-entrant melting behavior and the polymorphism of the solid phase. In simulations based on the machine learning potential we find evidence for continuous metallization in the liquid, as a first-order liquid-liquid transition is pre-empted by freezing. This suggests a smooth transition between insulating and metallic layers in giant gas planets, and reconciles existing discrepancies between experiments as a manifestation of supercritical behavior.
BibTeX
@inproceedings{cheng2020evidence,
  author = {Bingqing Cheng and Guglielmo Mazzola and Michele Ceriotti},
  title = {{Evidence for supercritical behavior of high-pressure liquid hydrogen}},
  booktitle = {Nature},
  year = {2020},
  doi = {10.1038/s41586-020-2677-y},
}
Figure