betaFeedback

WorksPhysics and Astronomy

The phase diagram of fluid hydrogen near melting from machine-learned interatomic potentials trained on quantum Monte Carlo

Abstract

Dissociation of molecular hydrogen increasing temperature or pressure may either occur gradually or abruptly, via a phase transition. Despite considerable experimental and theoretical efforts, the phase diagram of dense hydrogen at temperatures above solid melting is not precisely known. Here, we perform classical and quantum Path Integral molecular dynamics simulations of dense hydrogen using two machine-learned interatomic potentials both trained on quantum Monte Carlo energies and forces. The resulting phase diagram is constructed using state-of-the-art techniques. Both models show a first-order phase transition from the molecular to the atomic liquid terminating in critical points whose locations were determined using finite-size scaling techniques. Triple points, located at the highest temperature of the melting curve, are found at the intersection of the liquid-liquid lines with the melting curve and should lead to clear signatures in ramp compression experiments.

Introduction

Hydrogen exhibits a rich and not fully resolved phase diagram. Characterizing its behavior over a broad range of temperatures and pressures is important for understanding the structure and evolution of astrophysical systems such as stars, gas giants, and dense interstellar matter. Ever since the prediction of metallization by Wigner and Huntington [1], the observation of pressure-induced metallization has been a central objective in high-pressure physics. In the solid phase, experiments discovered a sequence of structural phase transitions [2] of the molecular crystal under static compression, with no clear signs of dissociation, whereas dynamic compression experiments at higher temperatures show metallic behavior in the fluid phase [3, 4, 5, 6].

Fluid hydrogen exists in two phases: a molecular insulating phase and an atomic metallic phase. For pressures less than 150 GPa and temperatures below 2500 K fluid hydrogen exists primarily as an insulating molecular liquid. Upon increasing pressure or temperature the molecules dissociate turning the system into an atomic fluid. However, since there is no difference in symmetry between fluid molecular hydrogen and fluid atomic hydrogen, whether the change happens via a crossover or a phase transition has been a long-standing question [7, 8, 9, 10, 11] since the work of Zeldovich and Landau [12].

The first evidence of a phase transition was provided by Density Functional Theory (DFT) based First-Principles Molecular Dynamics (FPMD) simulations in the NPT ensemble [13]. Since then, the first-order nature of the transition has been confirmed by many DFT-FPMD studies [9, 14, 15, 16] (see ref [7] for an extended bibliography). However, the location of the transition line is sensitive to the employed density functional as well as to nuclear quantum effects [17]. Simulation methods based on Quantum Monte Carlo (QMC) which are expected to represent the potential energy surface of the protons more accurately, such as the Coupled Electron-Ion Monte Carlo method (CEIMC) [7, 17, 18] and the QMC-MD [19], also supported the existence of a first-order transition. Although CEIMC and QMC-MD are significantly different methods and are based on different QMC wave functions, their predictions about the LLPT transition line are in reasonable agreement. Although hydrogen, with its simple electronic structure and the absence of core electrons, is a favorable system to apply First-Principles methods, studying phase transition and critical points at the ab-initio level is still quite challenging. Establishing the nature of the transition and investigating the occurrence of the critical point requires spanning a wide range of system sizes and simulation times. This investigation can be achieved today by employing Machine Learnt force fields trained on datasets of nuclear configurations with first-principle features (energy, forces and stress). In principle this preserves a first-principle accuracy in predicting new configurations but allows one to perform simulations of much larger systems for much longer times. Subsequently, Istas et al. [11] investigated the LLPT with finite-size scaling methods (FSS). Using a ML model trained on DFT-PBE, they found a clear transition line and estimated the critical point of that model. Recently, another ML model trained on DFT-PBE data [20], found supercritical behavior, i.e. a continuous dissociation with pressure with the location of the Widom line agreeing with the established DFT-PBE LLPT line.

The goal of the present study is to provide a more realistic estimate of the phase diagram of hydrogen in the region of the metal-insulator and molecular-atomic transitions. The previous studies have several major problems. The first requirement of such a calculation is to have an accurate proton-proton potential. Although a DFT functional can be accurate, the PBE functional is likely not [21]. Both the LLPT line and the melting line can vary considerably when changing functional and with respect to QMC predictions [22]. Secondly, a critical point can only be identified by accurate simulations comparing systems with different number of atoms so that one can see that the heat capacity and/or the compressibility diverge in the thermodynamic limit below the critical point but not above. Although such studies could be accomplished directly with QMC or another accurate electronic structure method, direct calculations for systems with thousands of protons in the region of pressure and temperature where critical fluctuations are present are very costly. Using ML potentials trained on accurate QMC energies and forces allow such finite-sized studies to be accomplished. Finally, the protons at these conditions have significant quantum motion, so estimates of its critical properties require path integral simulations in the pressure-temperature range of the LLPT.

In this work we perform molecular dynamics (MD) and path integral MD calculations (PIMD) using two different machine-learned (ML) force fields trained with QMC data to explore the phase diagram of liquid hydrogen in the region of 100-200 GPa and 1500-3000 K. This allows us to estimate the influence of the ML potential on the phase diagram and other properties. Both models predict a first-order phase transition from the molecular to the atomic liquid terminating in critical points. Our best estimate of the phase diagram with comparison to experimental reflectivity measurements as a fingerprint of molecular dissociation and metallization, is shown in Figure 1.

Figure 1. Phase diagram of hydrogen around the LLPT. Solid lines are predictions for quantum protons while the dot-dashed lines are for classical protons. Orange lines indicate the LLPT and melting transition from a molecular hcp lattice (Phase-I) of the M29 model while blue lines indicate the same for the D2508 model. The estimated critical points are shown as stars. Points show results of shock experiments at NIF (red squares, deuterium) [5], DAC measurements (green circles, hydrogen) [23] and (green squares, deuterium) [24]. These points indicate the threshold of the reflectivity R=0.3R = 0.3 associated with the insulator-metallic transition and the LLPT. The black dashed line is the experimental melting line of hydrogen [25] inferred from changes in the measured Raman spectra.

Methods

In this section we outline the methods we used to estimate the liquid-liquid transition in dense hydrogen, in particular how we locate the critical point using finite-size scaling and histogram reweighting.

Configuration selection

The machine learning models were trained on 13,800 configurations of 96 protons in periodic boundary conditions, covering pressures from 50 GPa to 200 GPa and temperatures from 1600 K to 4000 K. In order to guard against a bias for or against one of the phases the number of training and testing configurations were split evenly between the three phases: solid molecular hydrogen, liquid molecular hydrogen and liquid atomic hydrogen.

Quantum Monte Carlo Methods

We used Diffusion (DMC) and Variational (VMC) Quantum Monte Carlo methods to determine the electronic energy and forces of the configurations that went into training our machine learning models. We computed the electronic ground state energies and forces using the QMCPACK [26, 27] software. We assumed the trial wavefunction was of the Jastrow-Slater form which is accurate for many-body hydrogen in this range of density [28]. The Slater determinant was taken from a DFT calculation with the PBE functional. The 2-body Jastrow factor was optimized for each configuration. We assume the Born-Oppenheimer approximation for all our configurations. Electrons are in their ground state for temperatures and densities relevant to the LL critical point since the Fermi energy of the atomic liquid for densities near the critical density (0.45 Å−3^{-3}) is roughly 2×1052 \times10^{5} K. We can safely say that electronic excitations are negligible for the thermodynamic properties. We used the Chiesa et al. estimator [29] to compute the forces on the protons.

Machine-Learning models

We used message-passing graph neural network interatomic potentials MACE [30] and DeePMD [31] to run large scale MD simulations. In these models, configurations are mapped onto a set of learnable features, which in turn are mapped to the total energy. The features associated with a given atom depend not only on the coordinates of its neighbors, but also on their neighbor’s features. Feature construction proceeds recursively so that information at a larger distance is used.

The two models were trained on the same QMC dataset.2 We will refer to our particular MACE model as M29 and the DeePMD model as D2508. Both are state-of-the-art machine learning models: MACE is a good representative of an equivariant machine learning model and DeePMD is a state-of-the-art representative of an invariant machine learning model [33, 34]. MACE differentiates itself from other equivariant message-passing architectures by explicitly incorporating higher-body interactions within the message-passing process. The DeePMD model uses invariant features and incorporates higher-body effects by applying line-graph transformation to create higher order graphs whose vertices correspond to higher-body terms and then does message passing over that hierarchy [35]. Hyperparameters such as interaction cutoffs were tuned by minimizing energy and force errors of a test set. Since DMC energies are much more accurate than forces and dictate thermodynamic properties, we prioritized the energy root-mean-square-error (RMSE) with respect to the force RMSE.

Evaluated on the test set, the M29 model has a RMSE of energy 4.8 meV/atom and a RMSE of each force component of 129.9 meV/Å; the D2508 model has a RMSE of 5.3 meV/atom and 131.5 meV/Å respectively.

Determination of phase behavior

We investigated the phase diagram using molecular dynamics (MD) in the NPT ensemble. We employed the LAMMPS [36] software when considering classical protons and used LAMMPS built-in PIMD [37] and the i-PI software [38] for simulating quantum protons. Nuclear exchanges were neglected as they are irrelevant for the conditions under investigation.

Due to the remarkably high temperature melting lines predicted in recent studies [22, 39], at variance with earlier theoretical and experimental indications [14, 25, 40, 41], it is necessary to check that the observed transition is liquid-liquid and not solid-liquid. The absence of Bragg peaks in the structure factor, as illustrated in Figure 2(c), shows the liquid character of phases on both sides of the transition. See Table 1 for upper and lower bounds of the melting temperatures of the two models as a function of pressure.

PPTlM29T_l^{M29}TuM29T_u^{M29}TlD2508T_l^{D2508}TuD2508T_u^{D2508}TQM29T_Q^{M29}
15016451660156015701603
17517801790164016501634
20017001720169017001478
22516201625159016201358

Table 1. Estimates of the melting temperatures (in K) of classical and quantum hydrogen with the M29 and D2508 models vs. pressure (GPa). The upper (u) and lower (l) bounds are determined for classical hydrogen using the 2 phase method with 2048 atoms. The quantum estimates are from using a method similar to that in ref. [39]. Details are given in the SI. The bounds for the M29 model are consistent with those of the M18 model from ref. [39]. The D2508 model melts at about 100K lower temperature than does the M29 from the molecular liquid but only 15K lower from the atomic liquid.

Figure 2. (a) Instantaneous density of a MD trajectory as a function of time with the M29 potential for 576 atoms at 193 GPa and 1700 K. (b) The histogram of densities during the trajectory. (c) The structure factor averaged on the full trajectory does not indicate any broken translational order, consistent with a liquid phase. (d) The average radial distribution function for low (ρ<0.54\rho< 0.54 Å−3^{-3}, green dots) and high (ρ>0.54\rho> 0.54 Å−3^{-3}, red crosses) density configurations.

To establish the first-order nature of the LLPT transition, we looked at the discontinuity in the equation of state (EOS), and the divergence of the compressibility and the heat capacity, with increasing system size. Other properties accompanying this transition and further details of our NPT MD runs are provided in the supporting information (SI).

The first indication of a first-order transition is the observed bimodality in density fluctuations at constant pressure in the transition region. In Figure 2(a) we show the instantaneous density of a MD trajectory for a system with 576 classical protons as a function of time at 193 GPa and T=1700T=1700K for the M29 model. The density fluctuates between the molecular and atomic phases roughly every few ps. The minimal effect of hysteresis occurs at all temperatures, for both classical and quantum MD. The density distribution is bi-modal as shown in Figure 2(b).

This character change, from predominantly molecular to predominantly atomic, is indicated by the reduction of the molecular peak in the radial distribution function, as seen in Figure 2(d). The radial distribution function in the low-density molecular phase exhibits a sharp peak at roughly 0.7 Å, the characteristic bond-length of H2_2 molecules. This peak is significantly reduced for the higher-density atomic phase, indicating a much reduced molecular character.

The bi-modal behavior is not enough to differentiate a first-order transition from a crossover behavior. For this, we investigate the heat capacity, CpC_p, (computed from the enthalpy fluctuations) and the isothermal compressibility, κ\kappa, (computed from the volume fluctuations) with increasing system size as shown in Figure 3. In the transition region, these properties exhibit a strong size dependence for temperatures below the critical temperature TcT_c (Figure 3(a,Figure 3c)), while they are independent of the system size for T≥TcT \ge T_c (Figure 3(b,Figure 3d)). This is a clear signature of a first-order character of the transition below TcT_c, terminating in a critical point at TcT_c.

Heat capacity and isothermal compressibility for the M29 model

Figure 3. Heat capacity at constant pressure CpC_p and isothermal compressibility κ\kappa for the M29 model. The separate symbols designate the number of atoms. The vertical dashed lines show the transition pressure obtained from Eq. (1). Panel (a) shows a phase transition at 1800 K as CpC_p gets sharper and higher with system size. Panel (b) shows a crossover at 3000 K as CpC_p is independent of system size. Panels (c) and (d) show that the compressibility exhibits the same behavior as in panels (a) and (b) for CpC_p.

To estimate the location of the transition line we first fit the EOS (density versus pressure) along several isotherms and different system sizes to the phenomenological form [11, 42]:

ρ(P)=D2tanh⁡[α(P−Pt)]+κb(P−Pt)+ρt.(1)\rho(P) = \frac{D}{2}\tanh\left[\alpha(P-P_t)\right] + \kappa_b(P-P_t) + \rho_t . \tag*{(1)}

Some of the fit parameters are reported in Table 2. Note that the estimated relative density change across the LLPT at fixed temperature (D/ρtD/\rho_t) is quite small: it is at most 4.0% for M29 and 4.7% for D2508.

modelT (K)PtP_t (GPa)ρt\rho_t (Å−3^{-3})D (Å−3^{-3})
M29 Clas1700193.50.53910.0232
M29 Clas1800184.30.52710.0210
M29 Clas2000168.40.50530.0208
M29 Clas2100161.10.49470.0183
M29 H1600174.50.51680.0237
M29 H1700167.90.50800.0242
M29 H1800161.10.49730.0228
M29 H2000149.10.47980.0238
D2508 Clas1800194.10.54200.0259
D2508 Clas2100169.10.50750.0248
D2508 Clas2400148.10.47490.0183
D2508 H1800170.10.51150.0199
D2508 H2100149.50.48070.0244
D2508 H2400132.90.45370.0210

Table 2. Temperature, pressure, density and density jump along the LLPT. The average density from simulations of classical protons (Clas) and quantum hydrogen (H) at each temperature was fit using Eq. (1). Shown are PtP_t, ρt\rho_t and DD from those fits.

The behavior of the compressibility κ\kappa corroborates the first-order nature of the transition. As shown in the EOS (Figure 4), below the critical temperature (panel a), the compressibility diverges in the thermodynamic limit while above the critical point (panel b) it does not depend significantly on the system size.

Density versus pressure and number of classical protons for the two models

Figure 4. Density versus pressure and number of classical protons for the two models, D2508 (squares, higher pressure) and M29 (dots, lower pressure), at two different temperatures. (a) Both models show first-order phase transition behavior at 1800 K. (b) M29 shows a cross-over at 2500K, whereas it is nearly critical for D2508.

Along each isotherm T0T_0, we obtain the density, ρ(P)\rho(P), as a continuous function of pressure using histogram reweighting [43, 44]. This allows us to calculate the maximum compressibility

κ(N)=max⁡P(∂ρ∂P)T=T0.(2)\kappa(N) = \max_{P}\left(\frac{\partial\rho}{\partial P}\right)_{T=T_0}. \tag*{(2)}

To isolate the singular contribution at the transition line, κt(N)\kappa_t(N), we subtract a size and pressure independent background term, κb\kappa_b, calculated at pressures 8 to 15 GPa above critical point PcP_c using N=1200 and 2592 atoms.

As shown in the upper panel of Figure 5 the compressibility diverges linearly with respect to system size at 1800 K for M29 and at 2200 K for D2508. At temperatures approaching the critical point this agreement with linear extrapolation deteriorates, indicating the breakdown of the first-order nature of the transition for both models.

Figure 5. (a) Estimation of the dependence of the maximum compressibility κt\kappa_t for M29 (blue) and D2508(orange) versus system size below the critical temperature and (b) at the critical temperature. Points are estimates of the maximum compressibility from histogram reweighting for classical hydrogen. To extrapolate to the thermodynamic limit, straight dotted and dashed lines are shown as guides to the eye. In the lower plot, the two lines overlap making them difficult to distinguish.

The location of the critical point

Although these observables show a clear sign of a first-order phase transition, the location of the critical point requires more refined finite-size scaling analysis described below. Table 3 summarizes the predictions of the critical point based on different observables and extrapolation methods for both models.

M29D2508
Critical point
Ising fit (quantum)2031 ±\pm 20 K<br>145 ±\pm 1.2 GPa<br>0.480 Å−3^{-3}2470 ±\pm 9 K<br>127 ±\pm 0.5 GPa<br>0.467 Å−3^{-3}
Ising fit (classical)2095 ±\pm 9 K<br>160 ±\pm 0.7 GPa<br>0.483 Å−3^{-3}2478 ±\pm 7 K<br>143 ±\pm 0.5 GPa<br>0.461 Å−3^{-3}
κt(N)\kappa_t(N) scaling (classical)2100 ±\pm 200 K<br>160 ±\pm 10 GPa<br>0.484 Å−3^{-3}2500 ±\pm 200 K<br>140 ±\pm 10 GPa<br>0.453 Å−3^{-3}
Triple point
hydrogen1678 K<br>169.4 GPa1558 K<br>186 GPa
deuterium1735 K<br>177.7 GPa
classical1805 K<br>184.0 GPa1673 K<br>204 GPa

Table 3. Estimates of the critical point and triple point temperatures, pressures and densities for the two models from different methods.

While the transition lines of our ML models are close to each other, the critical point location exhibits model dependence:160 GPa at 2100 K for M29, in contrast to 140 GPa at 2500 K for D2508. Incorporating nuclear quantum effects further shifts the hydrogen transition line lower by approximately

30 GPa for both models.

We narrow the uncertainty in the location of the critical point by mapping the finite-size scaling onto the 3D universality class to obtain the correct scaling at the critical point. Universality classes are determined by a small number of features: the system dimensionality, the range of the interparticle forces, and the number of components in the order parameter. Since our ML models have short-range interactions, our system is in three spatial dimensions and the order parameter, the density, is a scalar quantity, we assume that LLPT critical point for fluid hydrogen is in the 3D Ising model universality class.

To map our transition to Ising, we replace the Ising order parameter, the average magnetization per spin, by the density fluctuation, i.e. s→Δρ≡(ρ−ρc)s \to\Delta\rho\equiv(\rho-\rho_c), where ρc\rho_c is the average critical density. Within the finite-size scaling assumption [45], the behavior of the distribution function pL(ρ,P)p_L(\rho,P) of a system of size LL in the immediate vicinity of the critical point is

pL(Δρ,ΔP)∼Lyp(ΔρLy,L/ξ)e−βVΔP(3)p_L(\Delta\rho,\Delta P) \sim L^{y}p(\Delta\rho L^{y},L/\xi)e^{-\beta V\Delta P} \tag*{(3)}

where ξ\xi is the correlation length of the density fluctuations, ΔP=P−Pc\Delta P=P-P_c, and p(⋅,⋅)p(\cdot,\cdot) is a universal scaling function. In the vicinity of the critical point, the average density then behaves as

⟨Δρ⟩=L−yf(βΔPL3−yρcy/3,L/ξ)(4)\langle\Delta\rho\rangle= L^{-y}f(\beta\Delta P L^{3-y}\rho_c^{y/3},L/\xi) \tag*{(4)}

where f(⋅,⋅)f(\cdot,\cdot) is a function resulting from the integration over the above distribution function. At the critical point, ΔP=0\Delta P=0, L/ξ→0L/\xi\to0. Using the Ising exponents with y=β/νy=\beta/\nu, β≃1/3\beta\simeq1/3, ν≃2/3\nu\simeq2/3 [45] and κt(N)∼∂⟨Δρ⟩∂P\kappa_t(N)\sim\frac{\partial\langle\Delta\rho\rangle}{\partial P}, we find the scaling behavior

κt(N)κt(N′)=(NN′)2/3\frac{\kappa_t(N)}{\kappa_t(N')}=\left(\frac{N}{N'}\right)^{2/3}

distinct from the linear dependence on NN characteristic of a first-order transition below TcT_c.

Figure 5 shows that at 2100K for M29 and 2500K for D2508, both models follow the scaling κt(N)∼N2/3\kappa_t(N)\sim N^{2/3} as expected in the vicinity of the critical point from the finite-size scaling analysis. This is especially visible in the M29 data for which we run one extra system size with N=2592N=2592 atoms. Similarly, above 2300K for M29 and 2800K for D2508, the extrapolated values vanishes, so that the change from molecular to atomic liquid becomes continuous.

To estimate the critical temperature and pressure more precisely without measuring very small density differences, we use the field mixing technique developed by Wilding et al. [46]. It is considered to be the most precise method to find the critical point of liquid-gas transition [43]: it has been applied to the Lennard-Jones liquid-vapor phase transition [47], and to the liquid-liquid phase transition

in water [48].

We compare the probability distribution of the order parameter for fluid hydrogen with that of the Ising model; in the critical region they converge to the same universal function [49]. The distribution of the magnetization of the 3D Ising model at the critical point can be approximated as [49]

pIsing(M)∝exp⁡(−(M2/M02−1)2(aM2/M02+c)),(5)p_{\mathrm{Ising}}(M) \propto\exp\left(-\left(M^{2}/M_{0}^{2}-1\right)^{2}\left(aM^{2}/M_{0}^{2}+c\right)\right), \tag*{(5)}

with a=0.158a = 0.158, c=0.776c = 0.776, and M0∝L−β/νM_{0} \propto L^{-\beta/\nu} with β/ν=0.518\beta/\nu= 0.518 [45, 50].

Wilding et al. [46] showed that to obtain an accurate comparison of the fluid critical fluctuations with the Ising model, identifying the magnetization with the fluid density suffers from large next to leading order effects. In a system that lacks symmetry between phases, an improved mapping of the order parameter is given by δm=m−⟨m⟩\delta m = m - \langle m\rangle, where m=ρ+sem = \rho+ se, ee being the potential energy per particle, ρ\rho the particle density and ss a scalar quantity controlling the degree of field mixing.

Using multiple histogram reweighting [43, 44, 51], we determine the corresponding probability distribution and the field mixing parameter ss of our hydrogen fluid containing NN protons, pH(δm∣P,T,N)p_{\mathrm{H}}(\delta m|P,T,N) for (P,T)(P,T) values close to the suspected critical point. Precisely at the critical point, pH(δm∣Pc,Tc,N)p_{\mathrm{H}}(\delta m|P_{\mathrm{c}},T_{\mathrm{c}},N) will approach the universal distribution, pIsingp_{\mathrm{Ising}}, in the limit of large systems. We therefore determine the critical pressure, Pc(L)P_{\mathrm{c}}(L), temperature, Tc(L)T_{\mathrm{c}}(L), for each system size N∼L3N \sim L^{3} by maximizing the overlap between (5) and pH(δm∣P,T,N)p_{\mathrm{H}}(\delta m|P,T,N).

Results are shown in Figure 6. The residual dependencies of the critical temperatures, Tc(L)T_{\mathrm{c}}(L), on the system sizes are shown in the lower panels of Figure 6. The apparent reduced critical temperature [46] converges as ∣Tc(L)−Tc(∞)∣∝L−(θ+1)/ν|T_{\mathrm{c}}(L)-T_{\mathrm{c}}(\infty)| \propto L^{-(\theta+1)/\nu} leading to Tc(∞)≃2478KT_{\mathrm{c}}(\infty) \simeq2478K for the D2508 model and to Tc(∞)≃2095KT_{\mathrm{c}}(\infty) \simeq2095K for the M29 model. The estimated critical pressures Pc(∞)P_{\mathrm{c}}(\infty) are then determined from the LLPT line at Tc(∞)T_{\mathrm{c}}(\infty).

Order-parameter distributions and critical-temperature finite-size scaling for the M29 and D2508 models

Figure 6. Procedures to estimate the LLPT critical point of hydrogen. (a) The order parameter distribution for the M29 model of classical hydrogen compared with the magnetization distribution of the 3D Ising model at criticality, given by Eq. (5). The field mixing value is fixed to s=−0.35s = -0.35 eV−1^{-1}Å−3^{-3} and is not sensitive to the system size; the proportionality constant M0∝L−β/νM_{0} \propto L^{-\beta/\nu} is fixed for one system size to match with the Ising distribution, β/ν=0.518\beta/\nu= 0.518 [50]. (bottom) Critical temperatures Tc(L)T_{\mathrm{c}}(L) for the M29 model (b) and the D2508 model (c) as a function of system size. Critical pressures and temperatures are extrapolated to the infinite system based on the critical exponents θ=0.54±0.05\theta= 0.54 \pm0.05 [52, 53] and ν=0.629\nu= 0.629 [52]. The blue points and continuous lines are for classical hydrogen, the red crosses and dashed lines for quantum hydrogen.

Quantum proton effects

We used PIMD to calculate the density fluctuations for quantum protons. We estimated the quantum critical point using a generalization of the histogram reweighting and finite-size scaling analysis described above: nuclear quantum effects do not modify the universality class at finite temperature. The changes

in the critical point are shown in the lower panels of Figure 6. We find that quantum effects move the transition line of hydrogen to lower pressures at a given temperature. The critical point is approximately (145 GPa, 2031 K) for the M29 model and (127 GPa, 2470 K) for the D2508 model as shown in Figure 1. Details of the procedures for histogram reweighting for path integrals are given in the SI.

Triple point

The intersections of the classical and quantum M18 melting lines [39] with the corresponding LLPT lines are used to determine the triple point, (T3,P3)(T_3, P_3), where the molecular hcp solid (S), molecular liquid (ML), and atomic liquid (AL) phases coexist. Results are given in Table 3. The slope of the solid H2H_2-liquid HH (S-AL) phase boundary, drawn in the phase diagram of Figure 1, follows from the Clausius-Clapeyron (CC) relations at the triple point among the three phases by using the slopes of the S-ML melting line, its volume change, the slope of the ML-AL line and its volume change. We have checked the melting temperatures of the two models with 2-phase studies of classical hydrogen with results consistent with the CC estimates: see Table 1. The location of the triple point is in agreement with the estimate in ref. [39] for the M18 model. Away from the critical point we estimated the quantum melting temperatures of hydrogen and deuterium monitoring the derivative of the melting temperature with respect to the ionic mass. Details are given in the SI.

Summary and Outlook

We predict the phase diagram of dense hydrogen using two machine-learning models as shown in Figure 1. We find a first-order phase transition in both models extending above the hcp molecular solid. But the two ML models differ in the precise location of the LLPT: at a fixed temperature, the difference between the two models is about 5% in the transition pressure and about 2.7% in the transition density. The difference in their critical temperatures is ≈18%\approx18\%.

The large difference in the critical point of the two ML models could arise from how the proton-proton interactions for distances larger than occur in the training data are treated in the ML models. Although size consistency is guaranteed in both ML models, use of ML energies and forces for systems larger than those used for training introduces an uncertainty which needs to be verified. We are currently investigating this by using data for larger systems in the training set. We note that smaller differences, ≈6%\approx6\%, are found in the melting temperatures of the two models: see Table 1.

In the SI we compare the current results with results from previous simulations using QMC and FP-DFT simulations [9, 10, 11, 13, 14, 15, 19, 20, 40, 54, 55, 56, 57, 58, 59] The difference between the current ML models is much less than with those of DFT predictions based on different XC-functionals. For example changing the DFT functional from PBE to vdW-DF1 changes the transitions/Widom lines by ∼70\sim70 GPa. The melting line of PBE-DFT is a factor of 2 lower than that obtain with the QMC trained model. We attribute the higher melting temperatures in the QMC trained models to the DFT models to the attractive dispersion forces of the molecular crystals that are not well approximated by the PBE-DFT functional. See ref. [39] for a comparison.

A consequence of the liquid-liquid first-order transition is the presence of a triple point on the melting curve. The dividing line between a metal and an insulator (MI) starts at the critical point, reaches the triple point and then continues along the melting line. At higher pressures the MI line follows the molecular melting line until either the molecular crystal becomes unstable or until it becomes metallic. The triple point could be a clear experimental indication of the LLPT that would be detected in ramp compression studies. The melting temperature will jump from increasing to decreasing with respect to pressure at the triple point caused by the higher density of the atomic liquid with respect to the molecular liquid. Without a first-order transition, the change under equilibrium conditions would be more gradual.

We expect that the LLPT for deuterium and its triple point will lie roughly halfway in between the hydrogen phase lines and the classical ones. We have tested this expectation using PIMD simulations of deuterium. See the SI for details.

Experimental observations of the metallic state in a dense fluid have been reported by dynamical compression experiments [5, 60, 61, 62, 63] and static compression methods heated by laser pulses [4, 23, 24, 64], yielding controversial results. Figure 1 shows comparison of the shock experiments of NIF [5] as well as dynamic DAC experiments [4, 23] that probed the pressure-temperature regime near the LLPT. We note that the NIF experiments do not directly measure temperature, the systems are dynamic on small inhomogeneous samples. The LLPT pressures is assigned to when the reflectivity reaches 30%. This corresponds to a minimum metallic conductivity of 2000 S/cm for bulk systems [65] indicative of the MI transition. In ref. [4] the temperature is inferred from a black-body radiation fit of the emitted light. Only integrated optical properties are observed. They are suggestive of the MI transition but by themselves cannot establish whether the transition is continuous or weakly first-order.

In conclusion, the ML models based on accurate Quantum Monte Carlo calculations predict a weak first-order transition between the insulating molecular liquid and the metallic atomic liquid. This transition is very near the melting transition. The sensitivity of the phase diagram to small changes in the proton-proton interaction leads to widely varying theoretical and experimental findings of the phase diagram. A reliable prediction requires an accurate and realistic treatment of molecular and atomic interactions, protonic quantum effects, and a careful examination of finite-size effects in the critical region and of the melting temperature.

We plan detailed exploration of thermodynamic properties in this region of the phase diagram. This will include experimental signatures such as Raman spectra [66] valuable for identifying the crystal structure. Direct comparison with experiment is of critical importance in validating the ML procedures and models. We also plan on extending these calculations to pressures between the triple point and the solid molecular –solid atomic transition.

Acknowledgments

Work by DC and SG was supported by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences (BES) under Award DE-SC0020177. HN was supported by NSFC under grant number 12504278 and 12611541151. YY was supported by the National Science Foundation (NSF) under grant number DMR-2532734. CP and IS were supported by the European Union - NextGenerationEU under the Italian Ministry of University and Research (MUR) projects PRIN2022-2022NRBLPT CUP E53D23001790006 and PRIN2022- P2022MC742PNRR, CUP E53D23018440001.

An award of computer time was provided by the INCITE program using resources of the Argonne Leadership Computing Facility, a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357. We also used the GRICAD infrastructure supported by the Grenoble research communities, Illinois Campus Cluster and the Delta advanced computing and data resource, supported by the National Science Foundation (Awards No. OCI-0725070, No. ACI-1238993 and award OAC 2005572) the State of Illinois, the University of Illinois at Urbana-Champaign, and its National Center for Supercomputing Applications. Delta is a joint effort of the University of Illinois Urbana-Champaign and its National Center for Supercomputing Applications. This work used Delta at National Center for Supercomputing Applications through allocation PHY240060 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program [67], which is supported by U.S. National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296. We acknowledge EuroHPC JU for awarding the project ID EUHPC E05 077 Extreme Access to the DCGP partition at Leonardo@CINECA, Italy and CINECA for providing computer time through the ISCRA-B project HP10B27XJC.

References

  1. [1]E Wigner and H B Huntington. On the possibility of a metallic modification of hydrogen. The Journal of Chemical Physics, 3(12):764–770, 1935.
  2. [2]Ho-kwang Mao and Russell J. Hemley. Ultrahigh-pressure transitions in solid hydrogen. Rev. Mod. Phys., 66:671–692, Apr 1994. doi: 10.1103/RevModPhys.66.671. URL https://link.aps.org/doi/10.1103/RevModPhys.66.671.DOI
  3. [3]Vitalii L. Ginzburg. What problems of physics and astrophysics seem now to be especially important and interesting (thirty years later, already on the verge of XXI century)? Phys.-Usp., 42(4):353, April 1999. ISSN 1063-7869. doi: 10.1070/PU1999v042n04ABEH000562. URL https://iopscience.iop.org/article/10.1070/PU1999v042n04ABEH000562/meta. Publisher: IOP Publishing.DOI
  4. [4]Mohamed Zaghoo, Ashkan Salamat, and Isaac F Silvera. Evidence of a first-order phase transition to metallic hydrogen. Physical Review B, 93(15):155128, 2016. doi: 10.1103/PhysRevB.93.155128.DOI
  5. [5]Peter M Celliers, Marius Millot, Stephanie Brygoo, R Stewart McWilliams, Dayne E Fratanduono, J Ryan Rygg, Alexander F Goncharov, Paul Loubeyre, Jon H Eggert, J Luc Peterson, et al. Insulator–metal transition in dense fluid deuterium. Science, 361(6403):677–682, 2018. doi: DOI:10.1126/science.aat0970.DOI
  6. [6]Eugene Gregoryanz, Cheng Ji, Philip Dalladay-Simpson, Bing Li, Ross T Howie, and Ho-Kwang Mao. Everything you always wanted to know about metallic hydrogen but were afraid to ask. Matter and Radiation at Extremes, 5(3), 2020.DOI
  7. [7]Michael Bonitz, Jan Vorberger, Mandy Bethkenhagen, Maximilian Böhme, David Ceperley, Alexey Filinov, Thomas Gawne, Frank Graziani, Gianluca Gregori, Paul Hamann, Stephanie Hansen, Markus Holzmann, S. X. Hu, Hanno Kählert, Valentin Karasiev, Uwe Kleinschmidt, Linda Kordts, Christopher Makait, Burkhard Militzer, Zhandos Moldabekov, Carlo Pierleoni, Martin Preising, Kushal Ramakrishna, Ronald Redmer, Sebastian Schwalbe, Pontus Svensson, and Tobias Dornheim. Toward first principles simulations of dense hydrogen. Physics of Plasmas, 31(11):110501, November 2024. ISSN 1070-664X, 1089-7674. doi: 10.1063/5.0219405. URL http://arxiv.org/abs/2405.10627. arXiv:2405.10627 [physics].
  8. [8]Bingqing Cheng, Guglielmo Mazzola, Chris J Pickard, and Michele Ceriotti. Evidence for supercritical behaviour of high-pressure liquid hydrogen. Nature, 585(7824):217–220, 2020.DOI
  9. [9]Valentin V Karasiev, Joshua Hinz, SX Hu, and Samuel B Trickey. On the liquid–liquid phase transition of dense hydrogen. Nature, 600(7889):E12–E14, 2021.DOI
  10. [10]Andrea Tirelli, Giacomo Tenti, Kousuke Nakano, and Sandro Sorella. High-pressure hydrogen by machine learning and quantum monte carlo. Physical Review B, 106(4):L041105, 2022.DOI
  11. [11]Mathieu Istas, Scott Jensen, Yubo Yang, Markus Holzmann, Carlo Pierleoni, and David M Ceperley. Liquid–liquid phase transition of hydrogen and its critical point: Analysis from ab initio simulation and a machine-learned potential. Physical Review E, 111(4):045307, 2025.
  12. [12]Ya. Zel’dovich and L. Landau. On the relation between the liquid and the gaseous states metals. Zhurnal Eksperimentalnoy I Teoreticheskoy Fiziki, 114:32, 1944.
  13. [13]Sandro Scandolo. Liquid–liquid phase transition in compressed hydrogen from first-principles simulations. Proceedings of the National Academy of Sciences, 100(6):3051–3053, 2003.pubmed.ncbi.nlm.nih.gov/12626753
  14. [14]Miguel A Morales, Carlo Pierleoni, Eric Schwegler, and David M Ceperley. Evidence for a first-order liquid–liquid transition in high-pressure hydrogen from ab initio simulations. Proceedings of the National Academy of Sciences, 107(29):12799–12803, 2010.DOI
  15. [15]Winfried Lorenzen, Bastian Holst, and Ronald Redmer. First-order liquid-liquid phase transition in dense hydrogen. Physical Review B—Condensed Matter and Materials Physics, 82(19):195107, 2010.DOI
  16. [16]GE Norman and IM Saitov. Critical point and mechanism of the fluid–fluid phase transition in warm dense hydrogen. Doklady physics, 62:294–298, 2017. doi: https://doi.org/10.1134/S1028335817060088.
  17. [17]Carlo Pierleoni, Miguel A Morales, Giovanni Rillo, Markus Holzmann, and David M Ceperley. Liquid–liquid phase transition in hydrogen by coupled electron–ion monte carlo simulations. Proceedings of the National Academy of Sciences, 113(18):4953–4957, 2016.DOI
  18. [18]Carlo Pierleoni, Markus Holzmann, and David M Ceperley. Local structure in dense hydrogen at the liquid–liquid phase transition by coupled electron–ion monte carlo. Contributions to Plasma Physics, 58(2-3):99–106, 2018. doi: https://doi.org/10.1002/ctpp.201700184. URL https://onlinelibrary.wiley.com/doi/abs/10.1002/ctpp.201700184.
  19. [19]Guglielmo Mazzola, Ravit Helled, and Sandro Sorella. Phase Diagram of Hydrogen and a Hydrogen-Helium Mixture at Planetary Conditions by Quantum Monte Carlo Simulations. Physical Review Letters, 120:025701, 2018. doi: 10.1103/PhysRevLett.120.025701. URL https://journals.aps.org/prl/pdf/10.1103/PhysRevLett.120.025701.DOI
  20. [20]Giacomo Tenti, Bastian Jäckl, Kousuke Nakano, Matthias Rupp, and Michele Casula. Hydrogen liquid-liquid transition from first principles and machine learning. Physical Review B, 112(10):104208, September 2025. ISSN 2469-9950, 2469-9969. doi: 10.1103/pbrk-3zgd. URL https://link.aps.org/doi/10.1103/pbrk-3zgd.DOI
  21. [21]Raymond C. Clay, Jeremy Mcminis, Jeffrey M. McMahon, Carlo Pierleoni, David M. Ceperley, and Miguel A. Morales. Benchmarking exchange-correlation functionals for hydrogen at high pressures using quantum monte carlo. Phys. Rev. B, 89:184106, May 2014. doi: 10.1103/PhysRevB.89.184106. URL https://link.aps.org/doi/10.1103/PhysRevB.89.184106.DOI
  22. [22]Hongwei Niu, Yubo Yang, Scott Jensen, Markus Holzmann, Carlo Pierleoni, David M Ceperley, et al. Stable solid molecular hydrogen above 900 k from a machine-learned potential trained with diffusion quantum monte carlo. Physical review letters, 130(7):076102, 2023.DOI
  23. [23]Vasily Dzyabura, Mohamed Zaghoo, and Isaac F. Silvera. Evidence of a liquid–liquid phase transition in hot dense hydrogen. Proceedings of the National Academy of Sciences, 110(20):8040–8044, 2013. doi: 10.1073/pnas.1300718110. URL https://www.pnas.org/doi/abs/10.1073/pnas.1300718110.DOI
  24. [24]Mohamed Zaghoo, Rachel J. Husband, and Isaac F. Silvera. Striking isotope effect on the metallization phase lines of liquid hydrogen and deuterium. Phys. Rev. B, 98:104102, Sep 2018. doi: 10.1103/PhysRevB.98.104102. URL https://link.aps.org/doi/10.1103/PhysRevB.98.104102.DOI
  25. [25]Chang-Sheng Zha, Hanyu Liu, John S Tse, and Russell J Hemley. Melting and High P-T Transitions of Hydrogen up to 300 GPa. Phys. Rev. Letts., 119:075302, 2017. doi: 10.1103/PhysRevLett.119.075302. URL https://journals.aps.org/prl/pdf/10.1103/PhysRevLett.119.075302.DOI
  26. [26]Jeongnim Kim, Andrew D Baczewski, Todd D. Beaudet, Anouar Benali, M. Chandler Bennett, Mark A. Berrill, Nick S. Blunt, Edgar Josué Landinez Borda, Michele Casula, David M. Ceperley, Simone Chiesa, Bryan K. Clark, Raymond C. Clay, Kris T. Delaney, Mark Dewing, Kenneth P. Esler, Hongxia Hao, Olle Heinonen, Paul R C Kent, Jaron T. Krogel, Ilkka Kylänpää, Ying Wai Li, M. Graham Lopez, Ye Luo, Fionn D. Malone, Richard M. Martin, Amrita Mathuriya, Jeremy McMinis, Cody A. Melton, Lubos Mitas, Miguel A. Morales, Eric Neuscamman, William D. Parker, Sergio D. Pineda Flores, Nichols A. Romero, Brenda M. Rubenstein, Jacqueline A R Shea, Hyeondeok Shin, Luke Shulenburger, Andreas F. Tillack, Joshua P. Townsend, Norm M. Tubman, Brett Van Der Goetz, Jordan E. Vincent, D ChangMo Yang, Yubo Yang, Shuai Zhang, and Luning Zhao. QMCPACK : an open source ab initio quantum Monte Carlo package for the electronic structure of atoms, molecules and solids. Journal of Physics: Condensed Matter, 30(19):195901, 5 2018. doi: 10.1088/1361-648X/aab9c3. URL https://iopscience.iop.org/article/10.1088/1361-648X/aab9c3.DOI
  27. [27]P. R. C. Kent, Abdul gani Annaberd iyev, Anouar Benali, M. Chandler Bennett, Edgar Josué Landinez Borda, Peter Doak, Hongxia Hao, Kenneth D. Jordan, Jaron T. Krogel, Ilkka Kylänpää, Joonho Lee, Ye Luo, Fionn D. Malone, Cody A. Melton, Lubos Mitas, Miguel A. Morales, Eric Neuscamman, Fernando A. Reboredo, Brenda Rubenstein, Kayahan Saritas, Shiv Upadhyay, Guangming Wang, Shuai Zhang, and Luning Zhao. QMCPACK: Advances in the development, efficiency, and application of auxiliary field and real-space variational and diffusion quantum Monte Carlo. The Journal of Chemical Physics, 152(17):174105, 5 2020. ISSN 0021-9606. doi: 10.1063/5.0004860. URL https://doi.org/10.1063/5.0004860http://aip.scitation.org/doi/10.1063/5.0004860.
  28. [28]Carlo Pierleoni, Kris T. Delaney, Miguel A. Morales, David M. Ceperley, and Markus Holzmann. Trial wave functions for high-pressure metallic hydrogen. Computer Physics Communications, 179(1):89–97, 2008. ISSN 0010-4655. doi: https://doi.org/10.1016/j.cpc.2008.01.041. URL https://www.sciencedirect.com/science/article/pii/S0010465508000210. Special issue based on the Conference on Computational Physics 2007.
  29. [29]Simone Chiesa, D. M. Ceperley, and Shiwei Zhang. Accurate, Efficient, and Simple Forces Computed with Quantum Monte Carlo Methods. Phys. Rev. Lett., 94(3):036404, 1 2005. ISSN 0031-9007. doi: 10.1103/PhysRevLett.94.036404. URL https://link.aps.org/doi/10.1103/PhysRevLett.94.036404.DOI
  30. [30]Ilyes Batatia, David P. Kovacs, Gregor Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields. Advances in Neural Information Processing Systems, 35:11423–11436, 2022.
  31. [31]Duo Zhang, Anyang Peng, Chun Cai, Wentao Li, Yuanchang Zhou, Jinzhe Zeng, Mingyu Guo, Chengqian Zhang, Bowen Li, Hong Jiang, Tong Zhu, Weile Jia, Linfeng Zhang, and Han Wang. A graph neural network for the era of large atomistic models. arXiv preprint arXiv:2506.01686, 2026. URL https://arxiv.org/abs/2506.01686.
  32. [32]David M Ceperley, Scott Jensen, Yubo Yang, Hongwei Niu, Carlo Pierleoni, and Markus Holzmann. Training models using forces computed by stochastic electronic structure methods. arXiv preprint arXiv:2310.15994, 2023.DOI
  33. [33]Anyang Peng, Chun Cai, Mingyu Guo, Duo Zhang, Chengqian Zhang, Wanrun Jiang, Yinan Wang, Antoine Loew, Chengkun Wu, Weinan E, et al. Lambench: a benchmark for large atomistic models. npj Computational Materials, 2026.arxiv.org/abs/2504.19578
  34. [34]Dávid Péter Kovács, Ilyes Batatia, Eszter Sára Arany, and Gábor Csányi. Evaluation of the MACE force field architecture: From medicinal chemistry to materials science. The Journal of Chemical Physics, 159(4):044118, July 2023. ISSN 0021-9606, 1089-7690. doi: 10.1063/5.0155322. URL https://pubs.aip.org/jcp/article/159/4/044118/2904837/Evaluation-of-the-MACE-force-field-architecture.DOI
  35. [35]Duo Zhang, Anyang Peng, Chun Cai, Wentao Li, Yuanchang Zhou, Jinzhe Zeng, Mingyu Guo, Chengqian Zhang, Bowen Li, Hong Jiang, et al. Graph neural network model for the era of large atomistic models. arXiv preprint arXiv:2506.01686, 2025.arxiv.org/abs/2506.01686
  36. [36]A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton. LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Comp. Phys. Comm., 271:108171, 2022. doi: 10.1016/j.cpc.2021.108171.DOI
  37. [37]Yifan Li, Axel Gomez, Kehan Cai, Chუნyi Zhang, Li Fu, Weile Jia, Yotam MY Feldman, Ofir Blumer, Jacob Higer, Barak Hirshberg, et al. fix pimd/langevin: An efficient implementation of path integral molecular dynamics in lammps. The Journal of Chemical Physics, 164(14), 2026.
  38. [38]Yair Litman, Venkat Kapil, Yotam M. Y. Feldman, Davide Tisi, Tomislav Begušić, Karen Fidanyan, Guillaume Fraux, Jacob Higer, Matthias Kellner, Tao E. Li, Eszter S. Pós, Elia Stocco, George Trinies, Barak Hirshberg, Mariana Rossi, and Michele Ceriotti. i-pi 3.0: A flexible and efficient framework for advanced atomistic simulations. The Journal of Chemical Physics, 161(6):062504, 08 2024. ISSN 0021-9606. doi: 10.1063/5.0215869. URL https://doi.org/10.1063/5.0215869.
  39. [39]Shubhhang Goswami, Scott Jensen, Yubo Yang, Markus Holzmann, Carlo Pierleoni, and David M Ceperley. High temperature melting of dense molecular hydrogen from machine-learning interatomic potentials trained on quantum monte carlo. The Journal of Chemical Physics, 162(5), 2025.DOI
  40. [40]Stanimir A. Bonev, Burkhard Militzer, and Giulia Galli. Ab initio simulations of dense liquid deuterium: Comparison with gas-gun shock-wave experiments. Phys. Rev. B, 69:014101, Jan 2004. doi: 10.1103/PhysRevB.69.014101. URL https://link.aps.org/doi/10.1103/PhysRevB.69.014101.DOI
  41. [41]Jie Luo, Tom Ichihiba, Yuan Ma, Yuan Liu, Jiaxiang Li, Ryo Maezono, Xin Zhong, Yu Xie, Hanyu Liu, Russell J. Hemley, and Yanming Ma. Quantum melting behavior of dense solid hydrogen from large-scale machine-learned potential simulations. Phys. Rev. B, 113:184116, May 2026. doi: 10.1103/47cf-cj2s. URL https://link.aps.org/doi/10.1103/47cf-cj2s.DOI
  42. [42]K. Binder and D. P. Landau. Finite-size scaling at first-order phase transitions. Physical Review B, 30(3):1477–1485, August 1984. ISSN 0163-1829. doi: 10.1103/PhysRevB.30.1477. URL https://link.aps.org/doi/10.1103/PhysRevB.30.1477.DOI
  43. [43]Kurt Binder, Dieter W Heermann, and K Binder. Monte Carlo simulation in statistical physics, volume 8. Springer, 2019.
  44. [44]Alan M Ferrenberg and Robert H Swendsen. Optimized monte carlo data analysis. Physical review letters, 63(12):1195, 1989.
  45. [45]K Binder. Finite size scaling analysis of ising model block distribution functions. Zeitschrift fur Physik B Condensed Matter, 43(2):119–140, 1981.
  46. [46]Nigel B Wilding. Simulation studies of fluid critical behaviour. Journal of Physics: Condensed Matter, 9(3):585–612, 1 1997. ISSN 0953-8984, 1361-648X. doi: 10.1088/0953-8984/9/3/002. URL https://iopscience.iop.org/article/10.1088/0953-8984/9/3/002.DOI
  47. [47]N.B. Wilding and K. Binder. Finite-size scaling for near-critical continuum fluids at constant pressure. Physica A: Statistical Mechanics and its Applications, 231(4):439–447, October 1996. ISSN 03784371. doi: 10.1016/0378-4371(96)00213-0. URL https://linkinghub.elsevier.com/retrieve/pii/0378437196002130.DOI
  48. [48]Pablo G. Debenedetti, Francesco Sciortino, and Gül H. Zerze. Second critical point in two realistic models of water. Science, 369(6501):289–292, July 2020. ISSN 0036-8075, 1095-9203. doi: 10.1126/science.abb9796. URL https://www.science.org/doi/10.1126/science.abb9796.DOI
  49. [49]M. M. Tsypin and H. W. J. Blöte. Probability distribution of the order parameter for the three-dimensional Ising-model universality class: A high-precision Monte Carlo study. Physical Review E, 62(1):73–76, July 2000. ISSN 1063-651X, 1095-3787. doi: 10.1103/PhysRevE.62.73. URL https://link.aps.org/doi/10.1103/PhysRevE.62.73.DOI
  50. [50]Alan M. Ferrenberg and D. P. Landau. Critical behavior of the three-dimensional ising model: A high-resolution monte carlo study. Phys. Rev. B, 44:5081–5091, Sep 1991. doi: 10.1103/PhysRevB.44.5081. URL https://link.aps.org/doi/10.1103/PhysRevB.44.5081.DOI
  51. [51]Mark Newman and Gerard Barkema. Monte carlo methods in statistical physics chapter 1-4. New York, USA, 1999.
  52. [52]Jing-Huei Chen, Michael E Fisher, and Bernie G Nickel. Unbiased estimation of corrections to scaling by partial differential approximants. Physical Review Letters, 48(9):630, 1982.DOI
  53. [53]Nigel B Wilding. Critical-point and coexistence-curve properties of the lennard-jones fluid: A finite-size scaling study. Physical Review E, 52(1):602, 1995.pubmed.ncbi.nlm.nih.gov/9963461
  54. [54]Isaac Tamblyn and Stanimir A. Bonev. Structure and phase boundaries of compressed liquid hydrogen. Phys. Rev. Lett., 104:065702, Feb 2010. doi: 10.1103/PhysRevLett.104. 065702. URL https://link.aps.org/doi/10.1103/PhysRevLett.104.065702. et al. Phase transition in a strongly non-ideal deuterium plasma generated by quasi-isentropical compression at megabar pressures. Physical Review Letters, 99(18):185001, 2007. doi: https://doi.org/10.1103/PhysRevLett.99.185001.
  55. [55]Miguel A. Morales, Jeffrey M. McMahon, Carlo Pierleoni, and David M. Ceperley. Nuclear quantum effects and nonlocal exchange-correlation functionals applied to liquid hydrogen at high pressure. Phys. Rev. Lett., 110:065702, Feb 2013. doi: 10.1103/PhysRevLett.110.065702. URL https://link.aps.org/doi/10.1103/PhysRevLett.110.065702.DOI
  56. [56]Hua Y Geng, Q Wu, Miriam Marqués, and Graeme J Ackland. Thermodynamic anomalies and three distinct liquid-liquid transitions in warm dense liquid hydrogen. Physical Review B, 100(13):134109, 2019.DOI
  57. [57]Joshua Hinz, Valentin V. Karasiev, S. X. Hu, Mohamed Zaghoo, Daniel Mejía-Rodríguez, S. B. Trickey, and L. Calderín. Fully consistent density functional theory determination of the insulator-metal transition boundary in warm dense hydrogen. Phys. Rev. Res., 2:032065(R), Sep 2020. doi: 10.1103/PhysRevResearch.2.032065. URL https://link.aps.org/doi/10.1103/PhysRevResearch.2.032065.DOI
  58. [58]Sebastiaan van de Bund, Heather Wiebe, and Graeme J. Ackland. Isotope quantum effects in the metallization transition in liquid hydrogen. Phys. Rev. Lett., 126:225701, Jun 2021. doi: 10.1103/PhysRevLett.126.225701.
  59. [59]Armin Bergermann, Lucas Kleindienst, and Ronald Redmer. Nonmetal-to-metal transition in liquid hydrogen using density functional theory and the heyd–scuseria–erznerhof exchange-correlation functional. The Journal of Chemical Physics, 161(23):234303, 12 2024. ISSN 0021-9606. doi: 10.1063/5.0241111. URL https://doi.org/10.1063/5.0241111.
  60. [60]ST Weir, AC Mitchell, and William J Nellis. Metallization of fluid molecular hydrogen at 140 GPa (1.4 mbar). Physical review letters, 76(11):1860, 1996. doi: https://doi.org/10.1103/PhysRevLett.76.1860.
  61. [61]VE Fortov, RI Il’kaev, VA Arinin, VV Burtzev, VA Golubev, IL Iosilevskiy, VV Khrustalev, AL Mikhailov, MA Mochalov, V Ya Ternovoi, et al. Phase transition in a strongly non-ideal deuterium plasma generated by quasi-isentropical compression at megabar pressures. Physical Review Letters, 99(18):185001, 2007. doi: https://doi.org/10.1103/PhysRevLett.99.185001.
  62. [62]MA Mochalov, RI Il’Kaev, VE Fortov, AL Mikhailov, AO Blikov, VA Ogorodnikov, VK Gryaznov, and IL Iosilevskii. Quasi-isentropic compressibility of a strongly nonideal deuterium plasma at pressures of up to 5500 gpa: Nonideality and degeneracy effects. Journal of Experimental and Theoretical Physics, 124:505–529, 2017.DOI
  63. [63]Marcus D Knudson, Michael P Desjarlais, Andreas Becker, Raymond W Lemke, KR Cochrane, Mark E Savage, David E Bliss, TR Mattsson, and Ronald Redmer. Direct observation of an abrupt insulator-to-metal transition in dense liquid deuterium. Science, 348(6242):1455–1460, 2015.
  64. [64]Shuqing Jiang, Nicholas Holtgrewe, Zachary M. Geballe, Sergey S. Lobanov, Mohammad F. Mahmood, R. Stewart McWilliams, and Alexander F. Goncharov. A Spectroscopic Study of the Insulator–Metal Transition in Liquid Hydrogen and Deuterium. Advanced Science, 7(2), 2020. ISSN 21983844. doi: 10.1002/advs.201901668.DOI
  65. [65]Giovanni Rillo, Miguel A. Morales, David M. Ceperley, and Carlo Pierleoni. Optical properties of high-pressure fluid hydrogen across molecular dissociation. Proceedings of the National Academy of Sciences, 116(20):9770–9774, 2019. doi: 10.1073/pnas.1818897116. URL https://www.pnas.org/doi/abs/10.1073/pnas.1818897116.DOI
  66. [66]Natarajan Subramanian, Alexander F. Goncharov, Viktor V. Struzhkin, Maddury Somayazulu, and Russell J. Hemley. Bonding changes in hot fluid hydrogen at megabar pressures. Proceedings of the National Academy of Sciences, 108(15):6014–6019, 2011. doi: 10.1073/pnas.1102760108. URL https://www.pnas.org/doi/abs/10.1073/pnas.1102760108.DOI

Paper details

vlnc:W_74pjgd66
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents