Influence of electronic entropy on Hellmann-Feynman forces in ab initio molecular dynamics with large temperature changes
Extracted source text · arXiv 2308.03963
% ****** Start of file apssamp.tex ******
%
% This file is part of the APS files in the REVTeX 4.2 distribution.
% Version 4.2a of REVTeX, December 2014
%
% Copyright (c) 2014 The American Physical Society.
%
% See the REVTeX 4 README file for restrictions and more information.
%
% TeX'ing this file requires that you have AMS-LaTeX 2.0 installed
% as well as the rest of the prerequisites for REVTeX 4.2
%
% See the REVTeX 4 README file
% It also requires running BibTeX. The commands are as follows:
%
% 1) latex apssamp.tex
% 2) bibtex apssamp
% 3) latex apssamp.tex
% 4) latex apssamp.tex
%
\documentclass[%
reprint,
%superscriptaddress,
%groupedaddress,
%unsortedaddress,
%runinaddress,
%frontmatterverbose,
%preprintnumbers,
%nofootinbib,
%nobibnotes,
%bibnotes,
amsmath,
amssymb,
%aps,
%pra,
prb,
%rmp,
%prstab,
%prstper,
%floatfix,
]{revtex4-2}
\usepackage{CJK}
\usepackage{multirow}
\usepackage{booktabs}
\usepackage{graphicx}% Include figure files
\usepackage{dcolumn}% Align table columns on decimal point
\usepackage{bm}% bold math
\usepackage{ulem}
\newcommand{\beginsupplement}{%
\setcounter{table}{0}
\renewcommand{\thetable}{S\arabic{table}}%
\setcounter{figure}{0}
\renewcommand{\thefigure}{S\arabic{figure}}%
}
%\usepackage{hyperref}% add hypertext capabilities
%\usepackage[mathlines]{lineno}% Enable numbering of text and display math
%\linenumbers\relax % Commence numbering lines
%\usepackage[showframe,%Uncomment any one of the following lines to test
%%scale=0.7, marginratio={1:1, 2:3}, ignoreall,% default settings
%%text={7in,10in},centering,
%%margin=1.5in,
%%total={6.5in,8.75in}, top=1.2in, left=0.9in, includefoot,
%%height=10in,a5paper,hmargin={3cm,0.8in},
%]{geometry}
\begin{document}
\begin{CJK*}{UTF8}{gbsn}
\title{Influence of electronic entropy on Hellmann-Feynman forces in {\it ab initio} molecular dynamics with large temperature changes}
\author{Ming Geng (耿明)$^{1,2}$}
\author{Chris E. Mohn$^{1,2,3}$}
\email{chris.mohn@kjemi.uio.no}
\affiliation{
$^{1}$Centre for Planetary Habitability (PHAB), University of Oslo, N-0315 Oslo, Norway
}
\affiliation{
$^{2}$Centre for Earth Evolution and Dynamics (CEED), University of Oslo, N-0315 Oslo, Norway
}
\affiliation{
$^{3}$Department of Chemistry and Center for Materials Science and Nanotechnology, University
of Oslo, Oslo 0371, Norway
}
\date{\today}% It is always \today, today,
% but any date may be explicitly specified
\begin{abstract}
The Z method is a popular atomistic simulation method for determining the melting temperature where a sequence of microcanonical molecular dynamics runs are carried out to target the lowest system energy where the solid always melts. Homogeneous melting at the limit of critical superheating, $T_{\text{h}}$, is accompanied by a drop in temperature as kinetic energy is converted to potential energy and the equilibrium melting temperature, $T_{\text{m}}$, can be calculated directly from the liquid state.
Implementation of the Z method interfaced with modern {\it ab initio} electronic structure packages use Hellmann-Feynman forces to propagate the ions in the microcanonical ensemble where the Mermin free energy plus the ionic kinetic energy is conserved. The electronic temperature, $T_{\text{el}}$, is therefore {\it kept fixed} along the trajectory which may introduce some spurious ion-electron interactions in molecular dynamics runs with large changes in temperatures such as often seen in homogeneous melting and freezing processes in the microcanonical ensemble.
We estimate possible systematic errors in the calculated melting temperature to choice of $T_{\text{el}}$ for two main mantle components, the wide band-gap insulators SiO$_2$ and CaSiO$_3$ at high pressure. Comparison of the calculated melting temperature from runs where the $T_{\text{el}}$ = $T_{\text{h}}$ and $T_{\text{el}}$ = $T_{\text{m}}$ - representing reasonable upper and lower boundaries respectively to choice of $T_{\text{el}}$ - shows that the difference in melting temperature is 200-300 K (3-5\% of the melting temperature) for our two test-systems.
Our results are in good agreement with previous large-size co-existence method and thermodynamic integration calculations, suggesting the CaSiO$_3$ and SiO$_2$ melts at around 6500 K (100 GPa) and 6000 K (160 GPa) respectively. The melting temperature decreases with increasing $T_{\text{el}}$ due to the increasing entropic stabilisation of the liquid and the systems melts typically about 3 times faster in molecular dynamics runs with $T_{\text{el}} = T_{\text{h}}$ compared to runs where $T_{\text{el}} = T_{\text{m}}$.
A careful choice of electron temperature in Born-Oppenheimer molecular dynamic simulations where the ions are propagated using Hellmann-Feynamn forces with the Mermin free energy + the ionic kinetic energy being conserved, is therefore essential for the critical evaluation of the Z method and in particular at very high temperatures.
\end{abstract}
\keywords{Suggested keywords}%Use showkeys class option if keyword
%display desired
\maketitle
\end{CJK*}
%\tableofcontents
\section{Introduction}
Melting and solidification processes are ubiquitous in condensed matter physics and in the evolution of terrestrial bodies including our own Earth. Triggered by extended defects, grain boundaries or open surfaces, equilibrium melting occurs spontaneously at $T_{\text{m}}$ when the free energy of the solid equals that of the liquid. Under certain conditions, however, a crystal can melt {\it homogeneously} at a much higher temperature, $T_{\text{h}}$, which represents the critical limit of superheating before melting is unavoidable~\cite{Belonoshko2006, Superheated2019}.
If a perfect periodic crystal in a molecular dynamics simulation is heated until $ T \approx T_{\text{h}}$, spontaneous fluctuations (nucleation precursors) will form transient defects and liquid nuclei which trigger an irreversible rapid nucleation growth and melting. Although homogeneous melting is rare in nature, shock-induced homogeneous melting has been demonstrated in a number of experiments (see for example~\cite {Russouw1985, Boyce1985, Cahn1986, Evans1986, Mo2018}).
The Z method explores the link between homogeneous and equilibrium melting using molecular dynamics simulations and has been widely used to estimate melting temperatures for different materials ~\cite{Clathrate2011, kaoliniteZ2013, SiC2020, Sliver2021} often at high temperatures and pressure where experiments are hazardous or impractical~\cite{AlumZAIMD2009, Belonoshko2012, 1TPa2015, Osmium2015, Vanadium256GPa2020, Zirconium2022}. A number of molecular dynamics simulations are launched in the microcanonical (NVE) ensemble at different initial temperatures, $T_{\text{ini}}$, to target the lowest total energy, $E_{\text{h}}$, where the solid always melts. When the system melts at $E_\text{h}(T_{\text{h}}$), the temperature decreases to the equilibrium melting temperature while the latent heat of melting gradually converts into potential energy. If we assume a linear variation of energy with temperature we can establish a relationship between $T_{\text{h}}$ and $T_{\text{m}}$ from the entropy of melting~\cite{Belonoshko2006, Braithwaite2019}:
\begin{equation}
\frac{T_{\text{h}}}{T_{\text{m}}} - 1 = \frac{\Delta S_{\text{m}}}{C_{\text{V}}}\label{eq1}
\end{equation}
where $C_{\text{V}}$ is the heat capacity at constant volume of the solid and $\Delta S_{\text{m}}$ is the entropy of melting. Eq.~\ref{eq1} can be approximated by ln2/3 assuming an ideal entropy of melting ($3k_\text{B}$ (per atom)) taken from a high-temperature limit of the Debye model and assuming that the heat capacity at constant volume is given by $k_\text{B}$ln2 (per atom) when $\Delta V_{\text{m}}/V \to 0 $ \cite{Stishov1973}.
Since the equilibrium melting temperature can be calculated at $T_{\text{h}}$ from a homogeneous melting triggered typically by small defects, quite small simulation boxes ($\sim$ 100 atoms) are in general sufficient to accurately calculate $T_{\text{h}}$ and hence $T_{\text{m}}$ from Eq.~\ref{eq1}~\cite{Davis2019}.
The use of small simulation boxes therefore makes the Z method a potentially attractive "low-cost" method for the accurate calculation of melting temperatures~\cite{Alfe2011, Gonzalez2016b, Davis2019}. By contrast, popular two-phase approaches to melting where the equilibrium melting process is mimicked by constructing a simulation box with a solid and a liquid phase in mechanical contact, often require large boxes to accommodate both solid and liquid phases and their interface~\cite{Hernandez2022, Brodholt2016}.
In addition, since $T_m$ can be calculated from Eq. \ref{eq1}, there is no need for the explicit calculation of free energies which sometimes hamper the precision of thermodynamic integration~\cite{UP-TILD2009, TU-TILD2015} and 2PT methods~\cite{2PT2003}, especially when the solid and liquid free energy curves have very similar steepness near $T_{\text{m}}$~\cite{Melting2017, Hernandez2022}.
In spite of these advantages, recent studies have unveiled some artificial features that may hamper the Z method for the accurate calculation of melting temperatures, particularly under extreme conditions~\cite{Hernandez2022, Alfe2011}. Since the waiting time required for the solid to melt diverges in the limit where $T_{\text{ini}} \to T_\text{h}$, a waiting time analysis is often carried out at different $T_{\text{ini}} > T_\text{h}$ to estimate $T_{\text{m}}$ from extrapolation to that of "infinite" waiting-time~\cite{Alfe2011}. This analysis, however, may still require extensive statistics for the precise calculation of melting temperatures.
Moreover, Born-Oppenheimer {\it ab initio} molecular dynamics (BOMD) is typically performed in the $NVE$ ensemble using Hellman-Faynman dynamics where the Fermi-Dirac electronic temperature is {\it kept fixed} in the MD simulation~\cite{ADV1992}. Although this implementation of the Z method ensures conserved dynamics~\cite{Wentzcovitch1992, Weinert1992}, large changes in temperature following melting and sometimes equilibration, may introduce systematic errors in $T_{\text{m}}$ since the electronic temperature is kept fixed~\cite{Hernandez2022}. That is, if a BOMD $NVE$ run with $E > E_{\text{h}}$ is launched with an $T_{\text{el}}$ {\it chosen near the melted liquid temperature} ($T_{\text{el}} \approx T_{\text{m}}$), then $T_{\text{el}}$ will be much lower than the temperature in the solid state (before melting). A too low electronic entropy may favour the stabilisation of the solid and prevent melting. This in turn will affect the estimated homogeneous melting temperature ($T_h$) and the "waiting time" for a solid to melt. The calculated equilibrium melting temperature may therefore be too high. On the other hand, if a BOMD run is launched with an electronic temperature {\it chosen near the solid temperature before melting} ($T_{\text{el}} \approx T_{\text{h}}$) a physically reasonable electronic-ionic interaction is ensured before melting. However, once the solid melts at constant volume, the temperature drops by 1 - $T_{\text{m}}/T_{\text{h}} \approx \text{ln}2/3$ and the electronic entropy will be much {\it higher} than the liquid temperature, which usually favours an entropic stabilisation of the liquid. Hence, the melting temperature may be too low. Therefore, since the choice of electronic entropy affect ion dynamics in processes undergoing large temperature drops, such as melting in the microcanonical ensemble, addressing the sensitivity in $T_{\text{m}}$ to the choice of $T_{\text{el}}$ is crucial in order to benchmark the Z method for the accurate calculating of melting temperatures.
In this work, we thus investigate the role of electronic entropy on homogeneous melting for two main abundant mineral components in the Earth's interior CaSiO$_3$ and SiO$_2$ at high temperatures.
Ca-perovskite is the third most abundant mineral in the lower mantle and a main component of basaltic lithologies constituting more than 20\% of recycled oceanic crust that is continuously being injected into the Earth's deep interior.
A strong preferential partitioning of radioactive heat-producing elements into CaSiO$_3$, such as U and Th, as well as key geochemical tracers, suggests that CaSiO$_3$ is the main storage minerals for many of these minority elements~\cite{Corgne2005, Tateno2018}. Tracking the distribution of CaSiO$_3$ in the lowermost mantle is therefore essential to understand the evolution of the solid Earth which in turn requires the thermodynamic conditions of Ca-perovskite melting. Motivated by this, a number of computational studies have calculated melting curves for pure CaSiO$_3$ to lowermost mantle conditions~\cite{Hernandez2022, Braithwaite2019}, but the agreement is not satisfactory. Here we attempt to contribute to tighten the constraints of CaSiO$_3$ melting at the lowermost mantle conditions.
Our second model system is SiO$_2$. The solid-liquid phase boundary of SiO$_2$ at ultrahigh pressure is critical to our understanding of not only the Earth's evolution but also the formation of many super-Earths~\cite{Das2020, Gonzalez2016a}. High-pressure silica melting may also play an important role in core dynamics, as it has been suggested that silica may have crystallized from a Si-saturated proto-core during a chemical exchange with a basal magma ocean~\cite{Hirose2017, Tronnes2019}.
In spite of a number of simulations and experimental results reported in the literature, the SiO$_2$ melting curve remains poorly constrained at very high pressure. Here we will attempt to contribute to resolving some of these outstanding discrepancies.
%Since ab initio MD simulations allow detailed insight into homogeneous melting processes we shall take the opportunity to investigate melting nucleation processes in these systems. Atomistic insight into the formation of defect formation triggering homogeneous melting remains poorly explored and is extremely difficult to capture experimentally.
\section{Theory}
The thermodynamic ensemble appropriate for the $Z$ method is the microcanonical ($NVE$) ensemble with the volume, $V$, number of species $N$ and the total system energy, $E$, kept fixed. The maximum energy along the solid branch of the isochore, $E_\text{h}$, is the same as the lowest energy along the liquid branch:
\begin{equation}
E_{\text{sol}}(V, T_{\text{h}}) = E_{\text{liq}}(V,T_{\text{m}}).\label{waitingtime}
\end{equation}
To locate $E_{\text{sol}}(V, T_{\text{h}})$ and hence $E_{\text{liq}}(V, T_{\text{m}})$, a sequence of $NVE$ MD runs are carried out with different initial temperatures. Since the waiting time for the solid to melt diverges when $T \to T_\text{h}$, the calculated melting temperature will always represents an upper bound to the "true" melting temperature.
To avoid extremely long MD runs in the vicinity of $T_\text{h}$, the melting temperature is calculated from an extrapolation of the distributions of waiting times using
\begin{equation}
\langle \tau \rangle^{-1/2} = A(T_{\text{liq}} - T_{\text{m}})
\label{Eqwait}
\end{equation}
where "A" is a parameter, $\tau$ is the waiting time for a solid to melt at a given total energy and $T_{\text{liq}}$ is the liquid temperature of the system after melting. Since $T_{\text{liq}} = T_{\text{m}}$, when $E = E_{\text{h}}$,
the melting temperature can be found at infinite waiting time i.e. at the point of intersection where $\langle \tau \rangle^{-1/2} = 0$.
We use {\it ab initio} Born-Oppenheimer MD to propagate the ions where the electronic energy is minimized at each step along the trajectory. Note that the usual Hellmann-Feynman forces do not conserve the total system energy, $E = U + K$ where $U$ is the internal DFT energy and $K$ is the kinetic energy of the ions. This is because the energy functional is non-variational with respect to changes in partial orbital occupancies along the MD trajectory.
To avoid additional contributions to the Hellmann-Feynman forces due to the variation in the band occupancies along the ionic trajectory the quantity "$K + \Omega$", rather than the total energy, is conserved in the MD run where $\Omega$ is the Mermin free-energy~\cite{Mermin1965, Wentzcovitch1992, Weinert1992}
\begin{equation}
\Omega = U-T_{\text{el}}S_{\text{el}}
\label{mermin}
\end{equation}
with
\begin{equation}
S_{\text{el}}=-k_{\text{B}}\sum _i[f_i \text{ln}{f_i}+(1-f_i)\text{ln}(1-f_i)].
\end{equation}
$S_{\text{el}}$ is the electronic entropy and $f_i$ is the electron occupancy of band $i$ calculated using Fermi-Dirac statistics:
\begin{equation}
f_i=F(\dfrac{\epsilon_i-E_{\text{fermi}}}{\sigma}),
\sum_{0}^{N_i}f_i=N
\end{equation}
where $F$ is the usual (Fermi-Dirac) smearing function, $\epsilon_i$ is the eigenvalue of band $i$, $E_{\text{fermi}}$ is the Fermi level and $\sigma$ = k$_\text{B}/T$ is the smearing broadening.
Ionic dynamics may be sensitive to the choice of electronic temperature when the ions are propagated using Hellmann-Feynman forces, in particular when the temperature changes are large such as during equilibration and melting in the $NVE$ ensemble. Electronic temperature may therefore affect the melting processes for semiconductors and insulators even though the fractional occupancies of the conduction bands, remain small during these temperature changes.
We can estimate the sensitivity in the calculated $T_{\text{m}}$ to choices of $T_{el}$ by comparison of the melting temperature calculated using $T_{\text{el}} \approx T_{\text{h}}$ with that calculated $T_{\text{el}} \approx T_{\text{m}}$. These choices of $T_{el}$ provide reasonable upper and lower bounds to the calculated melting temperature in the microcanonical ensemble when the Hellmann-Feynman forces are used for the ionic propagation with $K + \Omega$ being conserved.
If changes in $T_{\text{el}}$ can not be ignored, the usual conservation law including the Mermin functional in the form of Eq.~\ref{mermin} must be replaced, but the forces will then include contributions arising due to changes in the partial orbital occupancies along the microcanonical Born-Oppenheimer trajectory. It is important to note, however, that the time evolution of orbital occupancies due to large temperature changes may be only correctly described by the time-dependent Schr\"odinger equation.
A possible strategy to calculate $T_{\text{m}}$ using the Z method is to adjust $T_{\text{el}}$ along the ionic trajectory to match the average (ionic) temperature in the previous $N$ time-steps~\cite{Hernandez2022}. Although this "update scheme" does not ensure conserved dynamics, the average ensemble temperature before and after adjustment is in general expected to be small. In this approach, the time-dependent Schr\"ordinger equation is thus approximated by a sequence of BOMD NVE simulations.
\section{Computational details}
All BOMD simulations are performed with the Vienna {\it ab initio} simulation package (VASP) \cite{VASP93, VASP94}, using the projector augmented wave (PAW) method \cite{PAW94, PAW99}. We use the generalized gradient approximation (GGA) where the exchange-correlation contribution to the energy is parameterized using the PBE~\cite{PBE96} functional for SiO$_2$ and the AMO5 functional~\cite{Armiento2005} for CaSiO$_3$. The electronic configurations were: [He]2$s^2$2$p^4$ for O, [Ne]$3s^23p^2$ for Si and [He]$3s^3p^64^s2$ for Ca. The energy cutoff for the plane wave was 700 eV for SiO$_2$ and somewhat lower, 500 eV, for CaSiO$_3$ to compare directly with previous CaSiO$_3$ DFT studies~\cite{Hernandez2022, Braithwaite2019}.
In all runs, the atoms were placed at their ideal crystallographic sites i.e. the $1b$, $1a$ and $3d$ positions for Ca, Si and O atoms respectively of the $Pm3m$ space group (CaSiO$_3$). SiO$_2$ MD runs were started from the ideal cubic pyrite-type structure ($Pa\bar{3}$) optimized to target an equilibrium pressure $\sim$ 160 GPa. This is probably slightly below the stability field of the pyrite structured SiO$_2$ near the melting curve~\cite{Das2020}, but in order to compare directly with results in Ref.~\cite{Gonzalez2016a, Gonzalez2016b} we use pyrite rather than seifertite. The estimated melting point when pyrite is used as the crystal structure is only slightly lower compared to that found using seifertite.~\cite{Gonzalez2016a, Gonzalez2016b}.
For CaSiO$_3$ we use a cubic 135 atoms simulation box which is the same as that used in previous computational studies of CaSiO$_3$ melting~\cite{Braithwaite2019, Hernandez2022} allowing for a direct comparison with these studies. For SiO$_2$ the simulation box contained 96 atoms.
% COULD ADD THIS HERE OR STRATEGY
%In previous work, MD calculations are frequently launched with the atoms being located at their ideal equilibrium positions corresponding to the bottom of the well in the potential energy landscape~\cite{Hernandez2022, Braitwaite2019}[more]. As the system thermalice the high excess kinetic energy $\approx 2 \times T \rangle$ (solid) - drawn from a Maxwell--Boltzmann distribution - is rapidly converted to potential energy and the temperature drops. If $E < Eh$ then X equilibrate rapidly as a solid whereas if $E > Eh$ the systems will eventually melt (randomly).
Melting simulations are carried out in the $NVE$ ensemble with a timestep of 0.5 fs for SiO$_2$ and 1 ps for CaSiO$_3$. The smaller timestep for SiO$_2$ was chosen to minimize the energy fluctuation.
% This has been rewritten under
%Two types of simulation strategies were employed for SiO$_2$ simulations. One with fixed simulation time and the simulation run lasts at least 20 000 steps (10 ps), and another type runs until the system melted in order to carry out a waiting time analysis.
All runs used in the calculation of the waiting time were carried out until melting plus an additional 5-20 ps to calculate the average liquid temperature.
The waiting time analysis were performed based on between 8 and 20 different simulations at a given (E,V) where, in each run, the forces were taken from a Maxwell-Boltzmann distribution. Close to the equilibrium melting temperature we performed typically around 20 MD runs at a given ($E,V$) to ensure that sufficient statistics were collected in order to calculate $\langle \tau \rangle$ using Eq.~\ref{Eqwait}. Plots of the convergence of the estimated melting temperature with number of configurations are shown in supplementary information. This analysis shows that about 8-10 MD runs, for a given initial temperature, is sufficient to converge the melting temperature to less than 100 K. All MD calculations launched below $E_{\text{h}}$, lasted for at least 10 ps and close to $T_{\text{h}}$ the MD simulations typically ran for more than 100 ps.
For SiO$_2$, the $T_{\text{el}}$ are 6000 K, 7000 K and 8000 K where 6000 K is expected, after test-calculations, to lie close to $T_{\text{m}}$ whereas 8000 K will lie close to $T_{\text{h}}$. Similarly, for CaSiO$_3$ the waiting time analysis was carried out at $T_{\text{el}}$ = 6500 K and 9000 K which are expected to be close to $T_{\text{m}}$ and $T_{\text{h}}$ respectively.
To simulate melting with negligible contribution from the electronic entropy we used a Gaussian scheme~\cite{ADV1992} with a very low value of the smearing parameter (i.e. $\sigma_{\text{Gaussian}}$ = 0.03 eV).
The Gaussian smearing method is better designed to avoid instabilities arising from fluctuations in orbital occupancies at low values of $\sigma$ (low temperatures) which often hamper the Fermi-Dirac method during energy minimizations. The Gaussian smearing has the functional form $\frac{1}{2}(1-\text{erf}\dfrac{\epsilon-\mu}{\sigma})$ and the link between the two schemes is given by the ratios of the full width at half maximum, FWHM, as:
\begin{equation}
\frac{\text{FWHM}_{\text{FermiDirac}}}{\text{FWHM}_{\text{Gaussian}}}= \dfrac{\text{cosh}^{-1}(\sqrt{2})}{\sqrt{\text{ln}2}}.
\end{equation}
\section{Results and Discussion}
\subsection{Influence of electronic entropy on ionic dynamics}
\begin{figure*}
\includegraphics[width=\textwidth]{Fig2New4.pdf}
\caption{Temperature (red) and pressure (blue) evolutions along snapshots of {\it ab initio} MD trajectories for SiO$_2$ with different $T_{\text{el}}$. The thin lines show dumps every time-step whereas the thick lines are averaged properties over the previous 100 time-steps. All MD runs are carried out with the same initial temperature ($T_{\text{ini}}=18000$ K) where the atoms are distributed at their equilibrium lattice positions before we launch the MD run. A Gaussian smearing broadening scheme with $\sigma_{\text{Gaussian}}$ = 0.03 eV is used to represent simulations with a negligible contribution to electronic entropy on the ionic dynamics. The shaded area shows the melting process.}
\label{fig1}
\end{figure*}
In Fig.~\ref{fig1} we illustrate the sensitivity to changes in electronic temperature on the melting dynamics for silica where the initial ionic temperatures are {\it the same} in all runs.
In the extreme case where the electronic temperature is the same as the initial temperature in the MD runs ($T_{\text{el}} = T_{\text{ini}} $ = 18000 K), SiO$_2$ always melts rapidly and instantaneously in less than $ 0.5$ ps. On the contrary, in runs with negligible contribution from electronic entropy (i.e. with $\sigma_{\text{Gaussian}}$ = 0.03 eV), melting is rare and we observed only one incidence of melt-nucleation in all our 20 runs which lasted 10 ps each.
%.In this case, however, the systems quickly reverted to its initial solid state.
% put in example with CaSiO3 (no figure only text)?
In simulations with "intermediate" $T_{\text{el}}$, close to either the homogeneous or the equilibrium melting temperatures (i.e. with $T_{\text{el}} = 8000$ K or $T_{\text{el}}=6000$ K respectively), the system with $T_{\text{el}} = 6000$ K melted markedly slower compared to those with $T_{\text{el}} = 8000$ K and the waiting time was much longer. We found that the average waiting time was 2.7 ps when $T_{\text{el}}$ = 8000 K $\approx T_{\text{h}}$ and markedly longer by a factor of about 3 when $T_{\text{el}} = 6000 \approx T_{\text{m}}$. See also the discussion about distributions of the waiting time in the supplementary information. Interestingly, when $T_{\text{el}} = 6000$ K, we observe that the temperature sometimes drops markedly indicating possibly the formation of melt nuclei, but the system quickly reverted to its original (solid) state. This is seen as a small "bump" in the temperature/pressure evolution in Fig.~\ref{fig1} at about 2.0-2.5 ps. Solid-liquid "oscillations" are often seen in runs where the energy is close to the target homogeneous melting energy using small simulation boxes with large temperature fluctuations~\cite{Alfe2011} and will be discussed more in the supplementary information.
\begin{figure*}
\includegraphics[width=0.90\textwidth]{Graph3.pdf}
\caption{Two isochores with different electronic temperatures. The black curve is runs where $\sigma _{\text{Gaussian}}$ = 0.03 eV and the red curve is runs where $T_{\text{el}} = 8000 \approx T_{\text{m}}$. The initial temperatures are in the range 8000 K to 30000 K with the ions initially placed at their ideal lattice positions before we launched the MD simulations which ran for 20 ps each.}
\label{fig2}
\end{figure*}
The influence of electronic entropy on properties is also seen in Fig.~\ref{fig2} where two isochores with different electronic temperatures are drawn. Here, the initial ionic temperature in the MD runs is systematically increased from 8000 K to 30000 K. The isochore with an high electronic temperature i.e. close to $T_{\text{h}}$ (i.e. with $T_{\text{el}} = 8000$ K) deviates strongly from the one with negligible contributions from electronic entropy (i.e. $\sigma _{\text{Gaussian}}$ = 0.03 eV).
% What do you want to say here ? There is no volume work and the entropy of liquid is higher than that of solids (I have changed this below).
%With Gaussian smearing, since the electronic entropy %contribution is extremely small, the melting process mainly transforms the entropy of %ions ($TS$) and volume work ($pV$).
%, the melting process mainly transfers the entropy of ions ($TS$) and volume work ($pV$). Dont think we can say that there is no pressure changes if the electronic entropy was the main driveing factor to melting because the ions will also melt -> presse would increase.
%This temperature-pressure diagram shows that the difference in equilibrium temperature between the two isochores increases markedly with increasing temperature. Of note is that whereas the "tie-line" connecting the solid and liquid branches only shows a small change in temperature for Tel = 0 K, the tie-line decreases markedly when Tel = 8000 K. This again reflects the role of electronic entropy on melting: stabilisation of the liquid phase is partly due to contribution from electronic degrees of freedoms to total liquid entropy, and not the ions)
% Chris Comment: Figure 2B is almost the same as Figure 1.
%With increasing $T_{el}$, the equilibrium temperature decreases and then increases with a monotonic increase of equilibrium pressure (Fig.\ref{fig: Landing_T}B). The crossing in the pressure and temperature along with smearing also indicates the system favours staying in the solid when $T_{el}$ is low.
\subsection{Choice of electronic temperature in {\it ab initio} MD runs}
% MAINLY TALK ABOUT SMEARING EFFECT ON THE EQULIBRIUM IN THE MD. MAIN EVIDENCE: FIG1 and FIG2
\begin{figure*}
\includegraphics[width=0.94\textwidth]{Layout6.pdf}
\caption{Solid (blue) and liquid (red) branches in MD-$NVE$ runs for SiO$_2$ (left) and CaSiO$_3$ (right). For SiO$_2$, $T_{\text{el}}$ = 6000 K, 7000 K and 8000 K whereas for CaSiO$_3$, $T_\text{el}$ are 6500 K and 9000 K. In addition, we plot the result for CaSiO$_3$ with a Gaussian smearing scheme using $\sigma_{\text{Gaussian}} = 0.03$ eV. The homogeneous melting temperatures are calculated from an intersection of a linear extrapolation of the solid branch runs and a vertical line drawn from the equilibrium melting point (calculated using Eq. 3). The resulting homogeneous melting temperatures may therefore represent an upper bound to the "true" homogeneous melting temperature since the slope of the solid branch typically decreases near $T_{\text{h}}$~\cite{Davis2019,Hernandez2022}. Values of $T_{\text{h}}$ are reported in Table \ref{table}.}
\label{fig3}
\end{figure*}
\begin{figure*}
\includegraphics[width=0.95\textwidth]{Layout8.pdf}
\caption{The calculated melting temperature using Eq.~\ref{Eqwait} for different choices of electronic temperatures. For SiO$_2$, $T_{\text{el}}$ = 6000 K, 7000 K and 8000 K whereas for CaSiO$_3$, $T_\text{el}$ are 6500 K and 9000 K. In addition, we plot the waiting time for CaSiO$_3$ using a Gaussian smearing scheme with $\sigma_{\text{Gaussian}} = 0.03$ eV. The resulting estimated equilibrium melting temperatures are reported in Table 1~\ref{table} along with the standard error. The horizontal and vertical error bars reported are the mean errors of the waiting time and temperatures respectively.}\label{fig4}
\end{figure*}
The sensitivity in melting temperature to choice of $T_{\text{el}}$ illustrated above for SiO$_2$ and discussed in the supplementary information suggests that BOMD runs in the $NVE$ ensemble where the ions are propagated using the Hellmann-Feynman forces with $\Omega + K$ being conserved may introduce some errors due to changes in electron-ion interactions following large drops in temperature $ \approx$ ln2/3 $\times T_{\text{h}}$. This is seen in the Z plots and the waiting-time analysis in Fig.~\ref{fig3} and~\ref{fig4} as well as in Table~\ref{table} for both SiO$_2$ and CaSiO$_3$. When $T_{\text{el}}$ is kept fixed at some value near $T_{\text{m}}$, the calculated equilibrium melting temperature will be markedly higher compared to that if $T_{\text{el}}$ is close to the homogeneous melting temperature.
The calculated melting temperatures with $T_{\text{el}} \approx T_{\text{m}}$ at around 100 GPa (CaSiO$_3$) and 150 GPa (SiO$_2$) are about 300 K and 200 K higher respectively than those calculated with $T_{\text{el}} \approx T_{\text{h}}$. This corresponds roughly to 10\% of the temperature drop accompanying melting. We expect that these absolute errors increase with increasing melting temperature and pressure since the temperature drop accompanying melting increases with increasing $T_{\text{h}}$.
A key question is therefore: what is the best choice of electronic temperature in the MD runs to minimize the errors in the calculated $T_{\text{m}}$ when the ions are propagated using Hellmann-Feynmann forces with $\Omega + K$ being conserved?
If we choose an electronic temperature very close to $T_{\text{h}}$ in MD runs with $ E = E_{\text{h}}$, the electronic temperature is very close to the (ensemble) average temperature before the system eventually melts. Runs with $T_{\text{el}} \approx T_{\text{h}}$ therefore enables the accurate calculation of $T_{\text{h}}$ as well as the waiting time and implies that MD runs where $T_{\text{el}} < T_{\text{h}} $ (if, say $T_{\text{el}} = T_{\text{m}}$) give too high homogeneous melting temperatures.
%The accurate evaluation of the homogeneous melting temperature in MD NVE runs is thus given by choosing an electronic entropy that is close to the average solid temperature in runs along the solid branch of the isochore. % When $E = E_\text{h}}$ we therefore target $T_{\text{el}} = T_\text{h}}$.
%As seen in Table 1 and Fig.~\ref{ZwaitCaSiO3} a similar discrepancy in the choice of the electronic entropy is seen in the calculated {\it equlibrium} melting temperature. For SiO$_2$ and CaSiO$_3$ at $\sim$ 150 GPa and 100 GPa respectively, the calculated homogeneous melting temperature with $T_{\text{el}} \approx T_{\text{m}}$ is about 250 K higher compared to MD runs with $T_{\text{el} = T_\text{h}}$.
%These discrepancies in $ T_{\text{h}} $ to choice of electronic temperatures are marginally higher than those estimated for $T_{\text{m}}$.
% CHECK This can be explained by the drop in temperature accompanying melting which increases slightly with decreasing electronic entropy.
%Tel will be much lower than the average entropy along the solid branch and hence Th will be too low as seen in the T-E diagram (see e.g. Fig~\ref{ZwaitCaSiO3}).
%The cal in turn will "favour" the stabalization of the solid compared to the liquid and hence the calculated
%melting temperature will be too low.
% LATER/INTRODUCTION We can provide constraint on this error by analysing the sensitivity to melting to choize of T(z).
%Thus to accurately determine T$_h$ the Fermi-Dirac temperature should lie as close
%as possible to T$_h$ with E just below E$_m$.
%The choice of electronic entropy to accurately determine the equilibrium melting temperature is less evident.
Although $T_{\text{el}} \approx T_{\text{h}}$ (with $E \approx E_{\text{h}}$) enables the accurate calculation of $T_{\text{h}}$, the melting temperature may be severely underestimated. This is because $T_{\text{el}}$ is much larger than the liquid (ionic) temperature which, in general, favours an entropic stabilization of the liquid over the solid and hence the calculated melting temperature will be too low. %(assuming that T$_{\text{h}}$ is accurately determined).
%Better explained?: Although a reasonable electronic entropy is imposed in the MD run before melting when T$_{el} \langle T \rangle^{solid} T_h$, the electronic entropy will be too high in the liquid state following a large drop in temperature accompanying melting.
If we rather chose $T_{\text{el}} \approx T_{\text{m}}$ - which is the lowest temperature on the isochore and therefore represents a reasonable lower bound to choice of $T_{\text{el}}$ - the homogeneous melting temperature and possibly the equilibrium melting temperature will be {\it overestimated}. Calculations where $T_{\text{el}}$ is chosen to target either $T_{\text{m}}$ or T$_{\text{h}}$, therefore, provide reasonable {\it upper and lower bounds} respectively to the true melting temperature.
%This however long as the calculations of waiting time is sufficiently precise.
% Evidence for the precise estimate of T$_{\text{m}}$ from our waiting time analyses, can be illustrated in one of the runs (highlighted as a star symbol in Fig.~\ref{ZwaitCaSiO3} with T$_{\text{el}}$ = 9000 K). In this simulations, we did 10 MD runs, each with the same total energy and initial atomic positions but with different forces. The total energy from these runs was less than 0.1 eV from $E(T_h)$, estimated from the waiting time analysis. Only two of these runs melted in less than 100 ps. The liquid temperature of these two runs were 6408 K and 6368 K in very good agreement with the melting temperature calculated from the waiting time analysis (6437 K). [For SiO$_2$ we observed the same].
\subsection{CaSiO$_3$ melting}
%Recent two-phase calculations and thermodynamic integration using DFT, suggests that CaSiO$_3$ melts at around 7000 K at core-mantle boundary (CMB) conditions~\cite{Hernandez2022}, %which is about 2000 K higher than that of bridgmanite.
%(The 2PT method, on the other hand, predicts a much lower melting curve~\cite{Hernandez2022, Wilson2021} which is explained by the ideal-mixing entropy term which gives a too negative liquid free energy and hence a too low melting temperature.)
%The large discrepancy between two recent Z method calculations~\cite{Hernandez2022,Braithwaite2019} is surprising bearing in mind that the same functional to DFT and similar initial conditions for the MD run where used. Whereas Hernandez {\it et al}~\cite{Hernandez2022} found a melting temperature of around 6500 and 105 GPa, Braithwaite and Stixrude~\cite{Braithwaite2019} reported a melting point at the same density which was 1300 K lower. Possible explanations between this discrepancy such as the formation of glass - rather than liquid - following melting was investigated in Ref.~\cite{Hernandez2022}.
%We here explore the role of the electronic entropy on the calculated melting temperature of CaSiO$_3$.
%In Table 1 the reported melting temperatures for CaSiO$_3$ using the Z method at the same density (4.12 cm$^3$ g$^{-1}$ atom$^{-1}$) as that used in previous computational studies~\cite{Braithwaite2019,} together with values reported in previous computational studies~\cite{Hernandez2022, Braithwaite2019} is in good agreement with the Z method value in Ref.~\cite{Hernandez2022}.
%Melting temperatures from three simulations with different choices of electronic temperature was carried out using the waiting-time analysis to interpolate to infinite waiting time using Eq.~\ref{waitingtime}
%In the simulation with T$_{\text{el}}$ = 6500 K, the electronic temperature was chosen to lie close to the expected melting temperature whereas the run with T$_{\text{el}}$ = 9000 K was chosen to lie close to the homogeneous melting temperatures.
%We also report results using a Gaussian smearing with $\sigma = 0.03$ eV representing a run with negligible contribution from the electronic entropy on the melting temperature.
As shown in Table \ref{table}, our melting temperatures are in very good agreement with a previous Z method study~\cite{Hernandez2022} and also in excellent agreement with those from thermodynamic integration and two-phase calculations~\cite{Hernandez2022}. The inclusion of electronic entropy is essential for the accurate calculation of melting temperature for CaSiO$_3$, in agreement with that found for other wide band gap insulators such as MgO ~\cite{Alfe2005}. That is, the calculated $T_{\text{h}}$ and $T_{\text{m}}$ without contribution from the electronic entropy are about 1000 K and 650 K higher respectively than those calculated using $T_{\text{el}}$ = 6500 K. We find that that the melting temperature is around 6300 K (with $T_{\text{el}}$ = 9000 K) and 6600 K (with $T_{\text{el}}$ = 6500 K).
As discussed above, the discrepancy between the calculated melting temperature using $T_{\text{el}}$ = 6500 K with that calculated using $T_{\text{el}}$ = 9000 K is non-negligible for the accurate calculation of melting temperature. This places some constraints on the accuracy of the Z method interfaced with BOMD where $\Omega$ + K is conserved and the forces are propagated using the Hellmann-Feynman forces in the $NVE$ ensemble.
Choice of electronic entropy could therefore possibly explain the large discrepancy of more than 1000 K between our melting point and that of a recent Z method study~\cite{Braithwaite2019} which was carried out at the same thermodynamic conditions as here. We use the same exchange-correlation functional to DFT as that employed in Ref.~\cite{Braithwaite2019}, but are unable to reproduce their melting point unless we set $T_{\text{el}}$ = $T_{\text{ini}}$. Ref.~\cite{Braithwaite2019}, however, do not report values of $T_{\text{el}}$ so no firm conclusions can be drawn.
%Possible explanations between this discrepancy such as the formation of glass - rather than liquid - following melting was investigated in Ref.~\cite{Hernandez2022}.
% Glue this in the final sentence for SiO3.
%This difference in melting temperature help quantifying the constraints on the accuracy of the Z method for the accurate calculation of melting temperature using BOMD runs within the NVE ensemble where T$_{\text{el}}$ is kept fixed along the entire trajectory.
%(MOVE THIS TO BELOW/REPETITION: Although the total energy is conserved when forces are propagated using the usual Hellman-Feynman forces, large changes in the average temperature following equilibration or melting may give an incorrect coupling of electronic and ionic degrees of freedom in either the solid or the liquid states of the trajectory depending on choice of electronic entropy.)
% Please add the following required packages to your document preamble:
% \usepackage{multirow}
\begin{table*}[]
\setlength{\tabcolsep}{3pt}
\renewcommand{\arraystretch}{1.2}
\begin{tabular}{ccccccc}
\hline
\hline
\textbf{System} &
\textbf{Method} &
\textbf{$T_{el}$ (K)} &
\textbf{P (GPa)} &
\textbf{$T_h$ (K)} &
\textbf{$T_m$ (K)} &
\textbf{Ref.} \\ \hline
\multicolumn{1}{c|}{\multirow{9}{*}{\textbf{CaSiO$_3$}}} &
\multicolumn{1}{c|}{Z-Method} & \multicolumn{1}{c|}{9000} & \multicolumn{1}{c|}{103.7$\pm$2.6} & \multicolumn{1}{c|}{8873$\pm$42} & \multicolumn{1}{c|}{6410$\pm$21} & This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{6500} &
\multicolumn{1}{c|}{103.0$\pm$2.5} &
\multicolumn{1}{c|}{8652$\pm$25} &
\multicolumn{1}{c|}{6647$\pm$49} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{$\sim$0} &
\multicolumn{1}{c|}{105.2$\pm$3.0} &
\multicolumn{1}{c|}{9605$\pm$11} &
\multicolumn{1}{c|}{7258$\pm$38} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{adjust$^a$} &
\multicolumn{1}{c|}{102.6$\pm$2.7} &
\multicolumn{1}{c|}{8506} &
\multicolumn{1}{c|}{6517} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{adjust$^b$} &
\multicolumn{1}{c|}{105$\pm$3.3} &
\multicolumn{1}{c|}{8806} &
\multicolumn{1}{c|}{6493} &
Hernandez {\it et al.} 2022~\cite{Hernandez2022} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{not reported} &
\multicolumn{1}{c|}{103.0$\pm$0.2} &
\multicolumn{1}{c|}{7120} &
\multicolumn{1}{c|}{5200} &
Braithwaite {\it et al.} 2019~\cite{Braithwaite2019} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{large-size co-existence} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{$\approx$103} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{6582} &
Hernandez {\it et al.} 2022~\cite{Hernandez2022}\\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Thermodynamic Integration} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{$\approx$103} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{6433} &
Hernandez {\it et al.} 2022~\cite{Hernandez2022} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{2-Phase thermodynamics} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{$\approx$103} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{5420} &
Hernandez {\it et al.} 2022~\cite{Hernandez2022} \\ \hline
\multicolumn{1}{c|}{\multirow{9}{*}{\textbf{SiO$_2$}}}&
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{6000} &
\multicolumn{1}{c|}{164$\pm$4.3} &
\multicolumn{1}{c|}{8058$\pm$39} &
\multicolumn{1}{c|}{6110$\pm$60} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{7000} &
\multicolumn{1}{c|}{164$\pm$4.3} &
\multicolumn{1}{c|}{7899$\pm$30} &
\multicolumn{1}{c|}{6015$\pm$35} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{8000} &
\multicolumn{1}{c|}{166$\pm$4.3} &
\multicolumn{1}{c|}{7789$\pm$17} &
\multicolumn{1}{c|}{5884$\pm$21} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{adjust$^a$} &
\multicolumn{1}{c|}{169$\pm$4.3} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{6044} &
This work \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Coexist} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{153.8} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{5990} &
Benlonoshko {\it et al.} 1995~\cite{Belonoshko1995} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Coexist} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{157.6} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{5986} &
Usui {\it et al.} 2010~\cite{Usui2010}\\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Shock Experiment} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{157.0} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{5543} &
Millot {\it et al.} 2015~\cite{Millot2015} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{Z-Method} &
\multicolumn{1}{c|}{not reported} &
\multicolumn{1}{c|}{132.3} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{5852} &
Gonz\'alez-Cataldo {\it et al.} 2016~\cite{Gonzalez2016a} \\
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{DAC Experiment} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{117} &
\multicolumn{1}{c|}{} &
\multicolumn{1}{c|}{$\approx$6200} &
Andrault {\it et al.} 2022~\cite{Andrault2022} \\
\hline \hline
\end{tabular}
\caption{Calculated melting temperature for CaSiO$_3$ and SiO$_2$ using the Z method together with previous values reported in the literature at similar pressure (i.e. about 103 GPa for CaSiO$_3$ and 160 GPa for SiO$_2$).
In the simulation where $T_{\text{el}} \sim 0$ we used Gaussian smearing, $\sigma _{\text{gauss}}$ = 0.03 eV, for the partial occupancies of the one-electron orbitals. In the simulation from Ref.~\cite{Hernandez2022} where $T_{\text{el}}$ is labelled as "adjust$^b$", the Fermi-Dirac smearing was updated about every 1 ps along the MD trajectory to match the average temperature in the previous $\sim$ 1 ps, as discussed in Ref.~\cite{Hernandez2022}. In "adjust$^a$" we update the electronic entropy only once along the MD trajectory after about 25 ps of propagation in the liquid state.
%$^c$ is an average solid temperature after more than 140 ps of statistics.
}\label{table}
\end{table*}
% Th from Z.p, Tm from wait.p, tr. pressure from ZP.p
% From JGR paper
%103.0 ± 0.2 & 7120 (40) not provided 5200 (30) not provided PBEsol B&S (2019)
%105 ± 3.3 & 8806 6659 6493 updated PBEsol This study
%104.8 ± 2.6 & 8585 6589 not calc. updated PBE This study
%105.5 ± 2.8 & 8757 6505 not calc. 9000 PBE This study
\subsection{SiO$_2$ melting}
% THIS PART MAINLY TALK ABOUT THE MELTING POINT GET FROM OUR SIMULATION. MAIN Evidence Table I.
%SiO$_2$ is one of the most common components in the planet's interior. The solid-liquid phase boundary of SiO$_2$ at very high pressure consitent with those in the Earth's deep interiour and even beyond is important to understand the Earth's evolution and possibly the formation of super-Earth~\cite{Das2020}. High-pressure silica melting may also play an important role in the core dynamics, as it has been suggested that silica may have crystallized from a Si-saturated proto-core during a chemical exchange with a basal magma ocean\cite{Hirose2017, Tronnes2019}.
%SiO$_2$ melting is therefore important in order to understand planet evolution, but in spite of a number of simulations and experimental results, the $SiO_2$ melting curve is not so well constrained - in particular at high pressure/temperature consistent with those in the core's core and in Super Earths.
There are not many experimental studies of SiO$_2$ that report melting temperatures to very high pressure. Results from a recent high-pressure experimental study~\cite{Millot2015} was fitted to an equilibrium melting curve to about 500 GPa ($T_\text{m}(P)=1968.5+307.8\times P^{0.485}$) suggesting that SiO$_2$ melts at around 5540 K at 157 GPa. This is slightly lower than that calculated from a Z method simulation \cite{Gonzalez2016a}, where the melting temperature was estimated to be 5850 K at 132 GPa. By contrast, a recent Diamond Anvil Cell (DAC) experiment \cite{Andrault2022, Andrault2020}, suggests a markedly higher melting curve compared to those mentioned above and the Clapeyron slope is also much steeper in the pressure region 120-150 GPa compared to that reported by e.g. Millot {\it et al}~\cite{Millot2015}.
The melting curves from two molecular dynamics simulations \cite{Belonoshko1995, Usui2010} using two-phase co-existing methods are in overall good agreement with that from the DAC study~\cite{Andrault2022, Andrault2020}, but without the rapid change in the Clapeyron slope seen in the DAC experiment at around 120 GPa.
Our calculated melting temperatures reported in Table ~\ref{table} are in overall good agreement with previous computational predictions at similar pressures~\cite{Belonoshko1995, Usui2010}.
The calculated melting temperature reported using $T_{\text{el}}$ = 8000 K, for example, is 5776 K. This is only slightly lower compared to those of Refs.~\cite{Belonoshko1995} and ~\cite{Usui2010} which are 5990 K and 5986 K respectively.
Of note is that the good agreement with that reported by Usui and Tschuchia~\cite{Usui2010} may be fortuitous because a very small two-phase simulation box containing only 48 atoms for each phase was used in Ref.~\cite{Usui2010}. Such a small box has a boundary that is of similar size as the solid and liquid portions, and many runs are needed at a given (E,V) to precisely determine the melting points~\cite{Hong2016}. Our result also suggests that the predicted equilibrium melting curve reported from a shock experiment study~\cite{Millot2015} may be too low since it is assumed that stishovite is able to crystallize at the time-scale of the experiment and the melting curve is therefore drawn at the bottom of the liquid branch of the Hugoniot. If, however, stishovite is unable to crystallize at the time-scale of the shock experiment (i.e. within a few nanoseconds), the melting temperature reported in Ref.~\cite{Millot2015} may be underestimated as suggested by ~\cite{Shen2016}. This interpretation is consistent with our calculated melting point.
% From Jean-Alexis
%Millot considers the shock is over-heating the solid above the melting line on the low pressure side of the Hugoniot Z-shape. Therefore he made the melting curve pass through the bottom of the Z-shape, which represents the last point of the liquid Hugoniot.
%Shen et al., Nature Materials (2015), instead suggests the melting line should be at the top the of stishovite Hugoniot. The temperature drops at the bottom of the Z-shape is explained by the fact the stishovite does not have the time to crystallize in few nanoseconds due to the small Gibbs free energy difference between the solid and the liquid at these conditions.
As discussed above, the calculated melting temperature using $T_{\text{el}}$ = 8000 K is slightly less than 200 K lower than that calculated using $T_{\text{el}}\approx T_{\text{h}} \approx 6000$ K. This difference in the calculated melting temperature is of similar size compared to that found for CaSiO$_3$ at similar conditions and confirms that the Z method may be hampered by some artificial feature for the accurate calculations of melting temperature with the Mermin free energy + ionic kinetic energy being conserved along the Born-Oppenheimer MD-NVE trajectory with $T_{\text{el}}$ kept fixed.
% LATER/CONCULSION
%This difference in melting temperature help quantifying the constraints on the accuracy of the Z method for the accurate calculation of melting temperature using BOMD runs within the NVE ensemble where T$_{\text{el}}$ is kept fixed along the entire trajectory.
%(MOVE THIS TO BELOW/REPETITION: Although the total energy is conserved when forces are propagated using the usual Hellman-Feynman forces, large changes in the average temperature following equilibration or melting may give an incorrect coupling of electronic and ionic degrees of freedom in either the solid or the liquid states of the trajectory depending on choice of electronic entropy.)
%One of these is the choice of $T_{el}$ in NVE runs where melting is accompanied by large drops in temperature $ \approx (ln 2/3) \times T_h$. If $T_{el}$ is kept fixed
%Tel is therefore much lower than the average $\langle T \rangle$ along the solid trajectory
%at some value near T$_m$ the calculated equilibrium melting temperature will be markedly higher compared to %that if $T_{el}$ is close to the homogeneous melting temperature as seen in Fig.~\ref{ZwaitCaSiO3}.
%A waiting period before melting can be found along the simulation process.
% This is now in the waiting time analysis
%The RMSD curves stay at the level of starting position. With the lower initial temperature (14000 K), the crystal stays in solid after 10 ps of simulation, and the same crystal is fully molten after 4 ps when the initial temperature raised to 16000 K. In the case of 18000 K, the crystal rapidly melted after the start of the simulation. The melting process is shown as a temperature drop and pressure raise in the temperature and pressure profile. It clearly shows the melting process only happens when the initial temperature is large enough to overcome the superheating limit and the waiting time can be shortened by giving a higher initial temperature to the system.
% This is integrated in discussion
%Fig.\ref{fig:IniT18k} are simulations with the same initial temperature (18000K) and different values of T$_{el}$.
% Part of this is integrated in DISCUSSION
%When the broadening is really small (Gaussian smearing), the system will wait in the solid phase for long time for this particular initial temperature. The scenario occurs similarly when we use Fermi-Dirac smearing. With increasing the smearing broadening, the waiting time shrinking and also we can find the slope of RMSD curves become steeper which indicates a faster melting process.
%MOVE THIS TO RESULTS AND DISCUSSION (C. or E.)
%With the final equilibrium temperature, pressure and phase state of series simulations, an isochore can be generated as in Fig.\ref{fig:Landing_T}a.
% repetition
%The point transform from solid to liquid indicates the homogeneous melting point of the material. The Z shape isochore is also where the name of the method, Z-method, come from.
% we discuss this in discussion
%Meanwhile, the final equilibrium temperatures and pressures with different smearing broadening also differ because of the different electronic entropy contributions involved in the simulations.
% Dont quite get this:
%As it shows in Fig.\ref{fig:Landing_T}b, the equilibrium pressures increases more rapidly than temperature along with the increasing of smearing broadening regardless it finally equilibrate at solid or liquid phase.
% This is also repetition from discussion
%That means larger smearing broadening usually has a lower equilibrium temperature respect to the same pressure.
% I dont quite get this either:
%We can also find two groups of simulations equilibrate at obviously different temperatures which corresponds to the one that stays in solid and one transforms to liquid respectively. In the liquid branch, the pressure curve monotonically increases while the temperature curve diffuses crease then increases. This difference ilayates the pressure increase is not fully caused by thermal expansion of the melt.
% NOTE: I have integrated this in DISCUSSION
%The homogeneous melting temperature of the material determined by the Z shape isochore will also varies with the choose of smearing broadening. Higher broadening will find a relative lower melting temperature. Since the melting temperature in the simulation is the temperature of the nuclei and the smearing broadening reflex the temperature of electron. Low melting temperature ($T_{nuclei}=T_m$) found by high smearing broadening ($k_BT_{el}$) gives a decouple of nuclei and electron temperatures ($T_{nuclei} \neq T_{el}$). With Born-Oppenheimer approximation, these two temperatures do not need to be exactly the same in the simulation, but lower the difference between these two is still a factor to improve the accuracy of the simulation.
% NOTE: This is also in discussion now but the 18000 is not relevant.
%In our test case, the melting temperature difference caused by smearing broadening choosing from $8000k_B$ to $18000k_B$ can be around 500K lower near 165GPa. This is $\approx10\%$ of the melting temperature at this pressure.
% \subsection{Role of electronic entropy on waiting-time}
% MAIN EVIDENCE FIG 3 and FIG 4
% moved down
%Extrapolation to infinite waiting time using Eq.~\ref{Eqwait} is affected by electronic entropy and it is %therefore important to analyse the sensitivity in waiting time to the choice of T$_{\text{el}}$.
%since the waiting time is strongly sensitive to band occupancy.
%Increasing the electronic entropy will speed up the waiting time for melting as illustrated in Fig YYY.
% OLD TEXT NOW MOVED UP
%In Fig.~\ref{fig: IniT18k} we illustrate this for silica where the initial (ionic) temperature is the same in all runs but with different electronic entropies.
%In the case where the electronic entropy is the same as the initial (ionic) temperature (i.e. T = 18000 K), SiO$_2$ always melts rapidly and instantaneously ($< 0.5$ ps), independent of the initial velocities. In the runs where T$_{\text{el}} \approx T_{\text{h}} \approx $ 8000 K, the waiting time for melting was much longer and melting slower, whereas in calculations where T$_{el} \approx T_{\text{m}} \approx $ O K the system never melted even after more than 20 ps of propagation. In particular, for realistic choices of T$_{el}$ (i.e T$_{\text{el}}$ and T$_{\text{m}}$) the average waiting time was Y times longer when T$_{\text{el}}$ = 6000 K compared to 8000 K XX ps (for 6000 K). In Fig.~\ref{fig: Landing_T} (right figure) we show
%Similarly for CaSiO$_3$ the waiting time was typically about 3 times longer in runs where $T_{el} = T_m$ compared to those where $T_{el} = T_h$.
%With the final equilibrium temperature, pressure and phase state of series simulations, an isochore can be generated as in Fig.\ref{fig: Landing_T}a.
\subsection{Sensitivity to changes in electronic entropy on the waiting time analysis}
%Extrapolation to infinite waiting time using Eq.~\ref{Eqwait} is affected by electronic entropy and it is therefore important to analyse the sensitivity in waiting time to the choice of T$_{\text{el}}$.
%Since the band occupancy strongly influence the ionic dynamics the waiting time is sensitive to electronic entropy.
The waiting time for the solid to melt is correctly described if we chose $T_{\text{el}} \approx \langle T \rangle_{\text{sol}}$ after equilibration (i.e. if, for example, $T_{\text{el}} = T_{\text{h}}$ when $E = E_{\text{h}}$). However, since the melting temperature is in general underestimated with $T_{\text{el}} = T_{\text{h}}$, interpolation to infinite waiting time using Eq.~\ref{Eqwait} gives a too low equilibrium melting temperature when $\langle \tau \rangle $ is extrapolated to infinite waiting time.
However, if we use a lower $T_{\text{el}}$ (i.e. if we chose $T_{\text{el}} = \langle T \rangle _{\text{liq}}$) the waiting time for the system to melt will be too slow. This implies that $\langle \tau \rangle ^{-1/2}$ should be shifted to lower temperatures indicating that the melting temperature calculated from the intersection $\langle \tau \rangle^{-1/2}= 0$ will be overestimated. The estimated melting temperatures from a waiting time analysis with $T_{\text{el}} \approx T_{\text{m}}$ and $T_{\text{el}} \approx T_{\text{h}}$, therefore, provide reasonable upper and lower bounds respectively to the "true" equilibrium melting temperature. These differences ($\sim$ 200-300 K) are much larger than the standard errors from the waiting time analysis (reported in Table \ref{table}) which, in our case, are always less than 60 K. Convergence plots of the calculated melting temperatures using Eq. 3 with number of MD runs show that only a few tens of MD calculation are needed for the accurate calculation of melting temperatures (see supplementary information).
\begin{figure*}
\includegraphics[width=0.9\textwidth]{TwoContin.pdf}
\caption{Blue and red lines are continuation runs after about 40 ps of simulations (green lines). These continuation runs restarted with the same atomic positions and forces as the last step of the initial (green) run, but with different electronic temperatures. For CaSiO$_3$, the MD run was initially launched with $T_{\text{el}}$ = 9000 K (green line) and then restarted with $T_{el}$ = 6500 K (blue line). For SiO$_2$, we started with $T_{\text{el}}$ = 7000 K (green line) and then continued with $T_{\text{el}}$ = 6000 K (blue line) and 8000 K (red line).}\label{propEl}
\end{figure*}
To further understand the role of electronic entropy on the Hellmann-Feynman dynamics we can follow a similar strategy as in Ref.~\cite{Hernandez2022} by adjusting the electronic entropy along the MD trajectory to match the new average ionic temperature after transition to the liquid state.
We thus pick one of the MD runs with $T_{\text{el}} \approx T_{\text{h}}$ and a total energy which is marginally higher than $E_{\text{h}}$. Using CaSiO$_3$ as an example, we expect that the average liquid temperature is close to the equilibrium melting temperature estimated from Eq. 3.
Indeed, $\langle T \rangle$ after melting is 6408 K at $\sim$ 105 GPa, which is within the error-bars of $T_{\text{m}}$ (6410 K $\pm$ 21 K). The simulation was then restarted after 25 ps of propagation in the liquid state with a new electronic temperature chosen to match the average liquid temperature. In Fig.~\ref{propEl} (top panel) we show the temperature evolution of this restarted run (blue). The total energy increased by about 1\% accompanied by a temperature increase of about 100 K to 6517 K (labelled as "adjust" in Table 1). The $\langle T \rangle _{\text{liq}}$ from this relaunched MD run (blue line) is higher and lower respectively than the melting temperature reported in Table \ref{table} for $T_{\text{el}} = 9500$ K and $T_{\text{el}} = 6500$ K respectively. This confirms that $T_{\text{el}} \approx T_{\text{h}}$ and $T_{\text{el}} \approx T_{\text{m}}$ represent reasonable upper and lower bounds to the "true" melting temperature
%in agreement with those found from SLcoex (6582 K) and TI (6433 K) at similar pressures is excellent (see table 1).
% NB! We have to skip this since Tel is too low:
%Similarly, for SiO$_2$ when the run with $T_{\text{el}} \approx T_{\text{h}} = $ 8000 K was restarted with $T_{\text{el}}$ reset to 6000 K (see bottom panel in Fig.~\ref{propEl}), the liquid temperature increased by about 100 K which is lower and higher than those calculated with T$_{el}$ kept fixed at 6000 and 8000 K respectively.
%Although $\Omega + K$ is not conserved by this readjustment of the electronic entropy(i.e. the Mermin free energy increases by as much as 1\% (!) the ensemble temperature change by only about 100 K
Fig.~\ref{propEl} for SiO$_2$ also demonstrates that changes in electronic entropy can have a substantial impact on the ionic dynamics. Here we relaunch a simulation by changing the electronic temperatures from 7000 K (green line) to $T_{\text{el}}$ = 6000 K (blue line) or $T_{\text{el}}$ = 8000 K (red line). Whereas the MD run with $T_{\text{el}}$ = 6000 K {\it recrystallized} after about 15 ps of propagation in the liquid state the simulation with $T_{\text{el}}$ = 8000 K remained stable in the liquid phase until the run was terminated after more than 35 ps.
\section{Conclusions}
In this work, we used the Z method together with {\it ab initio} Born-Oppenheimer molecular dynamics to calculate the melting temperature for SiO$_2$ and CaSiO$_3$ at outer core and lower mantle conditions respectively with simulation boxes containing $\sim$ 100 atoms only. The calculated melting temperature for CaSiO$_3$ is in excellent agreement with results from previously reported two-phase co-existence calculations and thermodynamic integration and is substantially higher than that calculated in a previous {\it ab initio} study using the Z method~\cite{Braithwaite2019}. A possible explanation for this discrepancy has been discussed. The calculated melting temperature for SiO$_2$ is also in overall good agreement with previous computational work carried out at similar pressure and temperature suggesting that the estimated equilibrium melting curve reported in Ref.~\cite{Millot2015} may be too flat. %Advantages and challenges with the Z method as well as the nature of liquid nuclei that triggers homogenious melting has been discussed.
One of the great advantages with the Z method compared to other popular methods to melting, such as two-phase simulations and thermodynamic integration, is that the equilibrium melting temperature can be accurately calculated using small or modest-sized simulation boxes. This is because melting near $T_{\text{h}}$ is in general triggered by the formation of defects and small liquid clusters which can be embedded in a small simulation box~\cite{Davis2019}.
%Indeed, inspection of the MD trajectories show that small liquid nuclei form after homogeneous melting of the oxygen sublattice which initiates the "irreversible" and rapid homogeneous melting of the entire system.In contrast, direct simulation of equilibrium melting often involves extended surfaces or grain boundaries which require large simulation boxes to accommodate both phases including their interface.
%The observation of homogeneous melting of entire sublattices before melt nuclei is formed has to our knowledge not previously been observed computationally or experimentally and challenge the conventional atomistic model of homogeneous melting processes.
%(the growth-state is not caputured using small boxes but the growth of these does not hamper the calculated Tm)
%In contrast, direct simulation of equilibrium melting often involves extended surfaces or grain boundaries which require much large simulation boxes to accommodate both phases and their interface.
%Since solid-liquid co-existence is not possible in NVE runs, large box sizes is not needed to capture the nature of nucleation sites for the accurate calculation of melting temperatures.
Moreover, since the melting temperature is obtained from the relationship between $T_{\text{h}}$ and $T_{\text{m}}$ using Eq.\ref{eq1}, the Z method avoids the calculation of free energies which can be extremely tedious and expensive.
% We do not discuss this
%In addition we do not see any evidence for the formation of extended "frozen in" defects or glass-like states/clusters which can easily form in two-phase MD simulations when the liquid portion crystallizes near the equilibrium melting temperature~\cite{Hernandez2022} and hamper the accurate calculation of $T_{\text{m}}$.
%We did not see any such defects in the solid phase of SiO2 and CaSiO3 before melting - not even in runs with E = Eh + delta.
The Z method can easily be implemented and interfaced with popular {\it ab initio} simulation software such as VASP~\cite{VASP93, VASP94} and is embarrassingly parallelizable.
In spite of many appealing advantages compared to other methods to melting, the Z method appears to be hampered by some artificial features which needs to be much better addressed.
One of these, investigated here, is the choice of electronic entropy in Born-Oppenheimer MD simulation carried out in the microcanonical ensemble when $ K + \Omega $ is a conserved and hence $T_{\text{el}}$ is kept fixed along the Hellmann-Feynman trajectory. Since melting is accompanied by a large drop in temperature we need to quantify the errors introduced due to the choice of electronic temperature/entropy. We, therefore, compare the melting temperature calculated using $T_{\text{el}} \approx T_{\text{h}}$ and $T_{\text{el}} \approx T_{\text{m}}$ since these represent reasonable upper and lower bounds to the "true" melting temperature. For SiO$_2$ and CaSiO$_3$ at high pressure and temperature, the difference in the calculated $T_{\text{m}}$ with $T_{\text{el}} \approx T _{\text{h}}$ and $T_{\text{el}} \approx T _{\text{m}}$ is about 200-300 K. Although these discrepancies are only a few percent of the melting temperature, they are not negligible for the accurate calculation of melting temperature and are important to take into account for a critical assessment of the Z method implemented together with Born-Oppenheimer MD in the $NVE$ ensemble with a constant electronic temperature.
% Put these thoughts in computational theory paper:
%On the other hand, rapid melting makes it easier to identify %the waiting time of the simulation and hence increase the %precision of the waiting time analysis. This can help reduce computational cost when performing the waiting time analysis: Previous studies and our model systems both indicate that the waiting time follows Eq. YYY remarkable well. To precisely define a waiting-time in a simulation the simulation time after initial thermalisation (Ek sim Ep) should, in general, be longer than the waiting-time itself. Assuming little deviation from Eq XXX, accurate estimation of melting temperature can be done from an waiting time analysis where the average production time - after initial thermalization but before melting - is on the size of a few ps only. Although it is important to include values of $\tau ^{-1/2}$ close to T$_m$ to confirm Eq. XXX, the rapid growth in waiting time close to T$_m$ and the wide distributions of waiting times makes this computational challenging. The distributions of waiting times for CaSiO$_3$ with Energy E and shown in Fig YYY.
% this is now in Fig 1 + discussion.
%However, the homogeneous melting temperature get from the isochore depends strongly on the melting rate.
% This has now been discussed -> obsole
%The free energy of liquid and solid increase with the temperature, but the difference gap between these two energy curves also become narrower when it is close to the melting temperature. This will lead to a consequence that entropy contribution from the electrons can act as an initiator of the melting process. When two states are energetically really close to each other, no smearing in the SCF calculation will cause an electron occupation sloshing. DFT calculations of excited states have shown that this sloshing instability causes the jump over between two states of the system in the DFT simulations\cite{RN3745}. Thermal excitation usually can be quenched by smearing function.
\begin{acknowledgments}
We thank Prof. Dario Alf\`e for the thoughtful discussion. The Centre for Earth Evolution and Dynamics is funded by the Research Council of Norway through its Centre of Excellence program (Grant 223272). We acknowledge financial support from the Research Council of Norway through its Centres of Excellence scheme, project number 332523 (PHAB). Computational resources were provided by the Norwegian infrastructure for high-performance computing (NOTUR, Sigma-2, Grants NN9329K and NN2916K).
\end{acknowledgments}
% The \nocite command causes all entries in a bibliography to be printed out
% whether or not they are actually referenced in the text. This is appropriate
% for the sample file to show the different styles of references, but authors
% most likely will not want to use it.
\nocite{*}
\bibliography{apssamp}% Produces the bibliography via BibTeX.
\clearpage
\beginsupplement
\input{Supplementary}
\end{document}
%
% ****** End of file apssamp.tincrease with the temperature, but the difference gap between these two energy curves
\title{Influence of electronic entropy on Hellmann-Feynman forces in {\it ab initio} molecular dynamics with large temperature changes
\\ Supplementary Information}
\author{Ming Geng $^{1}$}
\author{Chris E. Mohn$^{1,2,}$}
\email{chrism@kjemi.uio.no}
\affiliation{
$^{1}$The Centre for Earth Evolution and Dynamics (CEED) and Centre for Planetary Habitability (PHAB), University of Oslo, N-0315 Oslo, Norway
}
\affiliation{
$^{2}$Department of Chemistry and Center for Materials Science and Nanotechnology, University of Oslo, Oslo 0371, Norway
}
\maketitle
\begin{figure*}
\includegraphics[width=0.9\textwidth]{Graph17.pdf}
\caption{Convergence of T$_{\text{m}}$ with number of MD runs for CaSiO$_3$. Each "set" of MD runs contains three different initial temperatures.}\label{ConvergenceCaSiO3}
\end{figure*}
\begin{figure*}
\includegraphics[width=0.9\textwidth]{Graph19.pdf}
\caption{Convergence of T$_{\text{m}}$ with number of MD runs for SiO$_2$. Each "set" of MD runs contains three different initial temperatures.}\label{ConvergenceSiO2}
\end{figure*}
\subsection{Distribution of waiting times and mean errors in the calculated melting temperature}
% MAIN EVIDENCE FIG9 and FIG10
The use of a waiting time analysis via Eq.~\ref{Eqwait} provides a powerful implementation of the Z method, but many MD runs is still needed for the precise calculation of $T_{\text{m}}$. In Figs. S1 and S2 we show the convergence of the melting temperature with number of sets where each "set" contains one MD run from each initial temperature. This shows that about 10 simulations of each set (about 30 MD runs in total) gives mean errors of less than 100 K. However many of these runs lasted for more than 200 ps and more MD runs would have been beneficial, in particular close to the melting temperature. Although the stochastic nature of melt nucleation and long tails in the waiting time distribution remains a challenge to the Z method~\cite{Alfe2011, Davis2019}, the emergence of machine learning could provide a powerful future approach to improve statistics~\cite{VASP93, VASP94}.
The stochastic nature of melting is illustrated as an inset in Fig. S3. Here we show the time evolution of a number of MD runs with the same total energy, $E$, which is slightly above $E_{\text{h}}$. As seen, some calculations melted fairly quickly ($<$ 10 ps) whereas others remained stable for more the 200 ps. Although we do not have sufficient statistics to capture the functional form of the distribution of $\tau $, analysis carried out by Davis {\it et al}~\cite{Davis2019} showed - using a classical Lennard-Jones potential to model melting of Ar - that the distribution of waiting times appears to deviate markedly from that of an exponential decay.
Although exponential decays describe many dynamical atomistic processes such as hopping events in solids~\cite{Mohn1CuI, MohnBi2O3}, the time taken from random fluctuations that trigger melting to when melting is observed will always take a certain time~\cite{Davis2019}. This suggests that $\tau$ deviates from an exponential decay at short times~\cite{Davis2019}.
In Fig.~\ref{disttauCaSiO3} and ~\ref{disttauSiO2}, we plot the distribution of the waiting times and possible (fitted) distributions
for CaSiO$_3$ and SiO$_2$ respectively. The plots confirm that the melting time is sensitive to initial temperature and electronic entropy as discussed above. The distributions of $ \tau $ with $T_{\text{el}}$ close to $T_{\text{m}}$ are wide and with standard deviations which are on the size of the average waiting time. For CaSiO$_3$ runs (with $T_{\text{ini}}$ = 20000 K and $T_{\text{el}}$ = 6500 K), for example, the mean waiting time is $\sim$ 50 ps with standard deviations of similar size when taken from a normal distribution as shown in Fig.~\ref{disttauCaSiO3}.
If we assume that the melting temperature can be precisely determined by Eq.~\ref{Eqwait}, a fruitful strategy to minimize the computational cost in a waiting time analysis would be to choose the total energies/initial temperatures such that $\langle \tau \rangle < 10$ ps and $\tau > \tau_{\text{melting}} \approx$ 1 ps. The upper bound ensures that we minimize lengthy MD runs, whereas the lower bound, $\tau _{\text{melting}}$, ensures that the waiting time is markedly longer than the time it takes from liquid
precursors is formed until the entire system has melted which, in our cases where rather small simulations boxes have been used, are typical $ \sim $ 1 ps. (see Fig.~\ref{fig1} in the main text). If the waiting time is shorter than the melting time, it is difficult to precisely identify the waiting time in a given run and hence the error bars in $\langle \tau \rangle ^{-1/2}$ are large (as seen in Fig. 4 in the main text at short waiting times).
% DISCUSS: how much time is needed to calculate a melting temperature - compare with SLUSCHI values.
\begin{figure*}
\includegraphics[width=0.9\textwidth]{Graph8.pdf}
% \includegraphics[width=0.9\textwidth]{Graph12.pdf}
\caption{Distribution of waiting times for melting and examples of $NVE$ MD runs of CaSiO$_3$ with $T_{\text{ini}}$ = 20000 K and $T_{\text{el}}$ = 6500 K. A normal-distribution is fitted to the distribution probability density. The inset figure shows five different seeds of the temperature evolution in an NVE MD runs with $E > E_{\text{h}}$ for CaSiO$_3$. In all runs the atoms were located at their equilibrium positions and the velocities were drawn from a Maxwell-Boltzmann distribution.}\label{disttauCaSiO3}
\end{figure*}
\begin{figure*}
\includegraphics[width=0.95\textwidth]{Layout7.pdf}
\caption{Distribution of waiting times for melting of SiO$_2$. A function is fitted to the distribution probability density. In the left figure we show distributions from runs with the same electronic temperature but different initial temperature and in the right one the initial temperatures are the same but the electronic temperatures are different. In all runs, the atoms were located at their equilibrium positions and the velocities were drawn from a Maxwell-Boltzmann distribution. }\label{disttauSiO2}
\label{fig:SiO2_waiting_distribution}
\end{figure*}
\subsection{Solid-liquid oscillations}
\begin{figure*}
\includegraphics[width=0.9\textwidth]{Recrystallize.pdf}
\caption{Temperature and pressure evolution in an SiO$_2$ MD run where T$_{\text{ini}}$ = 17000 K and $T_{\text{el}}$ = 8000 K. Transition between liquid and solids take place at 25 ps, 50 ps and 60 ps. Note the overlapping fluctuations in temperature/pressure along solid and liquid states.}\label{fig6}
\end{figure*}
Although solid-liquid transitions may be triggered by small changes in electronic entropy, refreezing can also occur spontaneously in $NVE$ runs when the temperature fluctuations are large. That is, when there is an overlap between liquid and solid temperature distributions as discussed in Ref.~\cite{Alfe2011}.
Fig.~\ref{fig6} illustrates solid-liquid oscillations for SiO$_2$. Here the temperature drops after about 20 ps as the system melts, followed by a 30 ps evolution in the liquid state before the system rapidly recrystallizes. After 10 ps in the solid state, it melts again and remains stable in the melt for more than 20 ps before the run is terminated.
%In passing it is interesting to note that no defects form in the solid phase following the liquid-solid transition.
Since, however, the temperature fluctuations scale as $1/\sqrt{N}$ (where $N$ is the number of atoms in the simulation box), the overlap between the solid and liquid temperatures decreases with increasing box size and hence the frequency of the solid-liquid alternation decreases. Oscillation on a time scale of nanoseconds or less is therefore usually only seen in MD runs carried out in small simulation boxes ($\sim$ 100 atoms).
%Dictated by the nature of the basin of attraction on the the potential energy suFanface (PES) associated with the solid - i.e. points on the pes that map to the solid crystalline phase following rapid quenches - the system will alternate between the solid and liquid states in the energy interval of solid-liquid co-existence (add Figure here).
An interesting implication of such solid-liquid oscillation is that it is possible to estimate the melting temperature by targeting the total energy in which the system spends an equal amount of time in the solid and liquid phases. In this case, the entropy of the two phases are equal and hence the melting temperature can be extracted directly from the average temperature in the liquid state~\cite{Alfe2011}.
However, in our case of SiO$_2$ melting at $\sim$ 160 GPa, extremely long runs - probably on the size of tens of nanoseconds - would be needed to collect sufficient statistics. In addition, analysis of all trajectories shows that liquid-solid transition is rare in $NVE$ MD runs in SiO$_2$, and for CaSiO$_3$ we did not observe any re-crystallization events even though the size of the simulation box size is modest (135 atoms). The emergence of machine learning with "on the fly" construction of force-fields along the MD trajectory could provide a powerful future implementation of the Z method to improve statistics. This would enable one to explore such oscillations on much longer time-scales than that possible with conventional BOMD simulations.
%apsrev4-2.bst 2019-01-14 (MD) hand-edited version of apsrev4-1.bst
%Control: key (0)
%Control: author (8) initials jnrlst
%Control: editor formatted (1) identically to author
%Control: production of article title (0) allowed
%Control: page (0) single
%Control: year (1) truncated
%Control: production of eprint (0) enabled
\providecommand{\noopsort}[1]{}\providecommand{\singleletter}[1]{#1}%
\begin{thebibliography}{65}%
\makeatletter
\providecommand \@ifxundefined [1]{%
\@ifx{#1\undefined}
}%
\providecommand \@ifnum [1]{%
\ifnum #1\expandafter \@firstoftwo
\else \expandafter \@secondoftwo
\fi
}%
\providecommand \@ifx [1]{%
\ifx #1\expandafter \@firstoftwo
\else \expandafter \@secondoftwo
\fi
}%
\providecommand \natexlab [1]{#1}%
\providecommand \enquote [1]{``#1''}%
\providecommand \bibnamefont [1]{#1}%
\providecommand \bibfnamefont [1]{#1}%
\providecommand \citenamefont [1]{#1}%
\providecommand \href@noop [0]{\@secondoftwo}%
\providecommand \href [0]{\begingroup \@sanitize@url \@href}%
\providecommand \@href[1]{\@@startlink{#1}\@@href}%
\providecommand \@@href[1]{\endgroup#1\@@endlink}%
\providecommand \@sanitize@url [0]{\catcode `\\12\catcode `\$12\catcode
`\&12\catcode `\#12\catcode `\^12\catcode `\_12\catcode `\%12\relax}%
\providecommand \@@startlink[1]{}%
\providecommand \@@endlink[0]{}%
\providecommand \url [0]{\begingroup\@sanitize@url \@url }%
\providecommand \@url [1]{\endgroup\@href {#1}{\urlprefix }}%
\providecommand \urlprefix [0]{URL }%
\providecommand \Eprint [0]{\href }%
\providecommand \doibase [0]{https://doi.org/}%
\providecommand \selectlanguage [0]{\@gobble}%
\providecommand \bibinfo [0]{\@secondoftwo}%
\providecommand \bibfield [0]{\@secondoftwo}%
\providecommand \translation [1]{[#1]}%
\providecommand \BibitemOpen [0]{}%
\providecommand \bibitemStop [0]{}%
\providecommand \bibitemNoStop [0]{.\EOS\space}%
\providecommand \EOS [0]{\spacefactor3000\relax}%
\providecommand \BibitemShut [1]{\csname bibitem#1\endcsname}%
\let\auto@bib@innerbib\@empty
%</preamble>
\bibitem [{\citenamefont {Belonoshko}\ \emph {et~al.}(2006)\citenamefont
{Belonoshko}, \citenamefont {Skorodumova}, \citenamefont {Rosengren},\ and\
\citenamefont {Johansson}}]{Belonoshko2006}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~B.}\ \bibnamefont
{Belonoshko}}, \bibinfo {author} {\bibfnamefont {N.~V.}\ \bibnamefont
{Skorodumova}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{Rosengren}},\ and\ \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont
{Johansson}},\ }\bibfield {title} {\bibinfo {title} {Melting and critical
superheating},\ }\href {https://doi.org/10.1103/PhysRevB.73.012201}
{\bibfield {journal} {\bibinfo {journal} {Physical Review B}\ }\textbf
{\bibinfo {volume} {73}},\ \bibinfo {pages} {012201} (\bibinfo {year}
{2006})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Olgu\'{i}n-Arias}\ \emph {et~al.}(2019)\citenamefont
{Olgu\'{i}n-Arias}, \citenamefont {Davis},\ and\ \citenamefont
{Gutierrez}}]{Superheated2019}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {V.}~\bibnamefont
{Olgu\'{i}n-Arias}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Davis}},\ and\ \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Gutierrez}},\ }\bibfield {title} {\bibinfo {title} {Extended correlations
in the critical superheated solid},\ }\bibfield {journal} {\bibinfo
{journal} {The Journal of Chemical Physics}\ }\textbf {\bibinfo {volume}
{151}},\ \href {https://doi.org/10.1063/1.5111527} {10.1063/1.5111527}
(\bibinfo {year} {2019})\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Rossouw}\ and\ \citenamefont
{Donnelly}(1985)}]{Russouw1985}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.~J.}\ \bibnamefont
{Rossouw}}\ and\ \bibinfo {author} {\bibfnamefont {S.~E.}\ \bibnamefont
{Donnelly}},\ }\bibfield {title} {\bibinfo {title} {Superheating of small
solid-argon bubbles in aluminum},\ }\href
{https://doi.org/10.1103/PhysRevLett.55.2960} {\bibfield {journal} {\bibinfo
{journal} {Phys. Rev. Lett.}\ }\textbf {\bibinfo {volume} {55}},\ \bibinfo
{pages} {2960} (\bibinfo {year} {1985})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Boyce}\ and\ \citenamefont
{Stutzmann}(1985)}]{Boyce1985}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.~B.}\ \bibnamefont
{Boyce}}\ and\ \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont
{Stutzmann}},\ }\bibfield {title} {\bibinfo {title} {{Orientational Ordering
and Melting of Molecular {H$_2$} in an $a$-Si Matrix: NMR Studies}},\ }\href
{https://doi.org/10.1103/PhysRevLett.54.562} {\bibfield {journal} {\bibinfo
{journal} {Physical Review Letters}\ }\textbf {\bibinfo {volume} {54}},\
\bibinfo {pages} {562} (\bibinfo {year} {1985})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Cahn}(1986)}]{Cahn1986}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {R.~W.}\ \bibnamefont
{Cahn}},\ }\bibfield {title} {\bibinfo {title} {Materials science: Melting
and the surface},\ }\href {https://doi.org/10.1038/323668a0} {\bibfield
{journal} {\bibinfo {journal} {Nature}\ }\textbf {\bibinfo {volume} {323}},\
\bibinfo {pages} {668} (\bibinfo {year} {1986})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Evans}\ and\ \citenamefont
{Mazey}(1986)}]{Evans1986}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.~H.}\ \bibnamefont
{Evans}}\ and\ \bibinfo {author} {\bibfnamefont {D.~J.}\ \bibnamefont
{Mazey}},\ }\bibfield {title} {\bibinfo {title} {Solid bubble formation in
titanium injected with krypton ions},\ }\href
{https://doi.org/10.1016/0022-3115(86)90004-8} {\bibfield {journal}
{\bibinfo {journal} {Journal of Nuclear Materials}\ }\textbf {\bibinfo
{volume} {138}},\ \bibinfo {pages} {176} (\bibinfo {year}
{1986})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Mo}\ \emph {et~al.}(2018)\citenamefont {Mo},
\citenamefont {Chen}, \citenamefont {Li}, \citenamefont {Dunning},
\citenamefont {Witte}, \citenamefont {Baldwin}, \citenamefont {Fletcher},
\citenamefont {Kim}, \citenamefont {Ng}, \citenamefont {Redmer},
\citenamefont {Reid}, \citenamefont {Shekhar}, \citenamefont {Shen},
\citenamefont {Shen}, \citenamefont {Sokolowski-Tinten}, \citenamefont
{Tsui}, \citenamefont {Wang}, \citenamefont {Zheng}, \citenamefont {Wang},\
and\ \citenamefont {Glenzer}}]{Mo2018}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {M.~Z.}\ \bibnamefont
{Mo}}, \bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Chen}}, \bibinfo
{author} {\bibfnamefont {R.~K.}\ \bibnamefont {Li}}, \bibinfo {author}
{\bibfnamefont {M.}~\bibnamefont {Dunning}}, \bibinfo {author} {\bibfnamefont
{B.~B.~L.}\ \bibnamefont {Witte}}, \bibinfo {author} {\bibfnamefont {J.~K.}\
\bibnamefont {Baldwin}}, \bibinfo {author} {\bibfnamefont {L.~B.}\
\bibnamefont {Fletcher}}, \bibinfo {author} {\bibfnamefont {J.~B.}\
\bibnamefont {Kim}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{Ng}}, \bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Redmer}}, \bibinfo
{author} {\bibfnamefont {A.~H.}\ \bibnamefont {Reid}}, \bibinfo {author}
{\bibfnamefont {P.}~\bibnamefont {Shekhar}}, \bibinfo {author} {\bibfnamefont
{X.~Z.}\ \bibnamefont {Shen}}, \bibinfo {author} {\bibfnamefont
{M.}~\bibnamefont {Shen}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont
{Sokolowski-Tinten}}, \bibinfo {author} {\bibfnamefont {Y.~Y.}\ \bibnamefont
{Tsui}}, \bibinfo {author} {\bibfnamefont {Y.~Q.}\ \bibnamefont {Wang}},
\bibinfo {author} {\bibfnamefont {Q.}~\bibnamefont {Zheng}}, \bibinfo
{author} {\bibfnamefont {X.~J.}\ \bibnamefont {Wang}},\ and\ \bibinfo
{author} {\bibfnamefont {S.~H.}\ \bibnamefont {Glenzer}},\ }\bibfield
{title} {\bibinfo {title} {Heterogeneous to homogeneous melting transition
visualized with ultrafast electron diffraction},\ }\href
{https://doi.org/10.1126/science.aar2058} {\bibfield {journal} {\bibinfo
{journal} {Science}\ }\textbf {\bibinfo {volume} {360}},\ \bibinfo {pages}
{1451} (\bibinfo {year} {2018})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Finney}\ and\ \citenamefont
{Rodger}(2011)}]{Clathrate2011}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~R.}\ \bibnamefont
{Finney}}\ and\ \bibinfo {author} {\bibfnamefont {P.~M.}\ \bibnamefont
{Rodger}},\ }\bibfield {title} {\bibinfo {title} {{Applying the Z method to
estimate temperatures of melting in structure II clathrate hydrates}},\
}\href {https://doi.org/10.1039/C1CP21919G} {\bibfield {journal} {\bibinfo
{journal} {Physical Chemistry Chemical Physics}\ }\textbf {\bibinfo {volume}
{13}},\ \bibinfo {pages} {19979} (\bibinfo {year} {2011})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Benazzouz}\ \emph {et~al.}(2013)\citenamefont
{Benazzouz}, \citenamefont {Zaoui},\ and\ \citenamefont
{Belonoshko}}]{kaoliniteZ2013}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {B.~K.}\ \bibnamefont
{Benazzouz}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Zaoui}},\
and\ \bibinfo {author} {\bibfnamefont {A.~B.}\ \bibnamefont {Belonoshko}},\
}\bibfield {title} {\bibinfo {title} {{Determination of the melting
temperature of kaolinite by means of the Z-method}},\ }\href
{https://doi.org/10.2138/am.2013.4379} {\bibfield {journal} {\bibinfo
{journal} {American Mineralogist}\ }\textbf {\bibinfo {volume} {98}},\
\bibinfo {pages} {1881} (\bibinfo {year} {2013})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Saiz}(2020)}]{SiC2020}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {F.}~\bibnamefont
{Saiz}},\ }\bibfield {title} {\bibinfo {title} {An ab initio study on liquid
silicon carbide},\ }\href {https://doi.org/10.1016/j.jpcs.2019.109204}
{\bibfield {journal} {\bibinfo {journal} {Journal of Physics and Chemistry
of Solids}\ }\textbf {\bibinfo {volume} {137}},\ \bibinfo {pages} {109204}
(\bibinfo {year} {2020})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Baty}\ \emph {et~al.}(2021)\citenamefont {Baty},
\citenamefont {Burakovsky},\ and\ \citenamefont {Errandonea}}]{Sliver2021}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.~R.}\ \bibnamefont
{Baty}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Burakovsky}},\
and\ \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Errandonea}},\
}\bibfield {title} {\bibinfo {title} {Ab initio phase diagram of silver},\
}\href {https://doi.org/10.1088/1361-648X/ac23fb} {\bibfield {journal}
{\bibinfo {journal} {Journal of Physics: Condensed Matter}\ }\textbf
{\bibinfo {volume} {33}},\ \bibinfo {pages} {485901} (\bibinfo {year}
{2021})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Bouchet}\ \emph {et~al.}(2009)\citenamefont
{Bouchet}, \citenamefont {Bottin}, \citenamefont {Jomard},\ and\
\citenamefont {Z\'erah}}]{AlumZAIMD2009}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Bouchet}}, \bibinfo {author} {\bibfnamefont {F.}~\bibnamefont {Bottin}},
\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Jomard}},\ and\ \bibinfo
{author} {\bibfnamefont {G.}~\bibnamefont {Z\'erah}},\ }\bibfield {title}
{\bibinfo {title} {Melting curve of aluminum up to 300 {GPa} obtained through
ab initio molecular dynamics simulations},\ }\href
{https://doi.org/10.1103/PhysRevB.80.094102} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {80}},\ \bibinfo
{pages} {094102} (\bibinfo {year} {2009})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Belonoshko}\ and\ \citenamefont
{Rosengren}(2012)}]{Belonoshko2012}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~B.}\ \bibnamefont
{Belonoshko}}\ and\ \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{Rosengren}},\ }\bibfield {title} {\bibinfo {title} {High-pressure melting
curve of platinum from ab initio z method},\ }\href
{https://doi.org/10.1103/PhysRevB.85.174104} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {85}},\ \bibinfo
{pages} {174104} (\bibinfo {year} {2012})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Geng}\ \emph {et~al.}(2015)\citenamefont {Geng},
\citenamefont {Hoffmann},\ and\ \citenamefont {Wu}}]{1TPa2015}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.~Y.}\ \bibnamefont
{Geng}}, \bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Hoffmann}},\
and\ \bibinfo {author} {\bibfnamefont {Q.}~\bibnamefont {Wu}},\ }\bibfield
{title} {\bibinfo {title} {Lattice stability and high-pressure melting
mechanism of dense hydrogen up to 1.5 tpa},\ }\href
{https://doi.org/10.1103/PhysRevB.92.104103} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {92}},\ \bibinfo
{pages} {104103} (\bibinfo {year} {2015})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Burakovsky}\ \emph {et~al.}(2015)\citenamefont
{Burakovsky}, \citenamefont {Burakovsky},\ and\ \citenamefont
{Preston}}]{Osmium2015}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {L.}~\bibnamefont
{Burakovsky}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont
{Burakovsky}},\ and\ \bibinfo {author} {\bibfnamefont {D.~L.}\ \bibnamefont
{Preston}},\ }\bibfield {title} {\bibinfo {title} {Ab initio melting curve
of osmium},\ }\href {https://doi.org/10.1103/PhysRevB.92.174105} {\bibfield
{journal} {\bibinfo {journal} {Physical Review B}\ }\textbf {\bibinfo
{volume} {92}},\ \bibinfo {pages} {174105} (\bibinfo {year}
{2015})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Zhang}\ \emph {et~al.}(2020)\citenamefont {Zhang},
\citenamefont {Tan}, \citenamefont {Geng}, \citenamefont {Salke},
\citenamefont {Gao}, \citenamefont {Li}, \citenamefont {Sekine},
\citenamefont {Wang}, \citenamefont {Greenberg}, \citenamefont {Prakapenka},\
and\ \citenamefont {Lin}}]{Vanadium256GPa2020}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont
{Zhang}}, \bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Tan}}, \bibinfo
{author} {\bibfnamefont {H.~Y.}\ \bibnamefont {Geng}}, \bibinfo {author}
{\bibfnamefont {N.~P.}\ \bibnamefont {Salke}}, \bibinfo {author}
{\bibfnamefont {Z.}~\bibnamefont {Gao}}, \bibinfo {author} {\bibfnamefont
{J.}~\bibnamefont {Li}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont
{Sekine}}, \bibinfo {author} {\bibfnamefont {Q.}~\bibnamefont {Wang}},
\bibinfo {author} {\bibfnamefont {E.}~\bibnamefont {Greenberg}}, \bibinfo
{author} {\bibfnamefont {V.~B.}\ \bibnamefont {Prakapenka}},\ and\ \bibinfo
{author} {\bibfnamefont {J.-F.}\ \bibnamefont {Lin}},\ }\bibfield {title}
{\bibinfo {title} {Melting curve of vanadium up to 256 gpa: Consistency
between experiments and theory},\ }\href
{https://doi.org/10.1103/PhysRevB.102.214104} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {102}},\ \bibinfo
{pages} {214104} (\bibinfo {year} {2020})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Minakov}\ \emph {et~al.}(2022)\citenamefont
{Minakov}, \citenamefont {Paramonov}, \citenamefont {Demyanov}, \citenamefont
{Fokin},\ and\ \citenamefont {Levashov}}]{Zirconium2022}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.~V.}\ \bibnamefont
{Minakov}}, \bibinfo {author} {\bibfnamefont {M.~A.}\ \bibnamefont
{Paramonov}}, \bibinfo {author} {\bibfnamefont {G.~S.}\ \bibnamefont
{Demyanov}}, \bibinfo {author} {\bibfnamefont {V.~B.}\ \bibnamefont
{Fokin}},\ and\ \bibinfo {author} {\bibfnamefont {P.~R.}\ \bibnamefont
{Levashov}},\ }\bibfield {title} {\bibinfo {title} {Ab initio calculation of
hafnium and zirconium melting curves via the lindemann criterion},\ }\href
{https://doi.org/10.1103/PhysRevB.106.214105} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {106}},\ \bibinfo
{pages} {214105} (\bibinfo {year} {2022})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Braithwaite}\ and\ \citenamefont
{Stixrude}(2019)}]{Braithwaite2019}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Braithwaite}}\ and\ \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont
{Stixrude}},\ }\bibfield {title} {\bibinfo {title} {Melting of {CaSiO$_3$}
perovskite at high pressure},\ }\href {https://doi.org/10.1029/2018GL081805}
{\bibfield {journal} {\bibinfo {journal} {Geophysical Research Letters}\
}\textbf {\bibinfo {volume} {46}},\ \bibinfo {pages} {2037} (\bibinfo {year}
{2019})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Stishov}\ \emph {et~al.}(1973)\citenamefont
{Stishov}, \citenamefont {Makarenko}, \citenamefont {Ivanov},\ and\
\citenamefont {Nikolaenko}}]{Stishov1973}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.~M.}\ \bibnamefont
{Stishov}}, \bibinfo {author} {\bibfnamefont {I.~N.}\ \bibnamefont
{Makarenko}}, \bibinfo {author} {\bibfnamefont {V.~A.}\ \bibnamefont
{Ivanov}},\ and\ \bibinfo {author} {\bibfnamefont {A.~M.}\ \bibnamefont
{Nikolaenko}},\ }\bibfield {title} {\bibinfo {title} {On the entropy of
melting},\ }\href {https://doi.org/10.1016/0375-9601(73)90485-4} {\bibfield
{journal} {\bibinfo {journal} {Physics Letters A}\ }\textbf {\bibinfo
{volume} {45}},\ \bibinfo {pages} {18} (\bibinfo {year} {1973})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Davis}\ \emph {et~al.}(2019)\citenamefont {Davis},
\citenamefont {~},\ and\ \citenamefont {J.}}]{Davis2019}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Davis}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {~},
\bibfnamefont {Loyola}},\ and\ \bibinfo {author} {\bibfnamefont
{P.}~\bibnamefont {J.}},\ }\bibfield {title} {\bibinfo {title} {Bayesian
statistical modeling of microcanonical melting times at the superheated
regime},\ }\href@noop {} {\bibfield {journal} {\bibinfo {journal} {Physica
A}\ }\textbf {\bibinfo {volume} {515}},\ \bibinfo {pages} {546} (\bibinfo
{year} {2019})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Alf\'e}\ \emph {et~al.}(2011)\citenamefont {Alf\'e},
\citenamefont {Cazorla},\ and\ \citenamefont {Gillan}}]{Alfe2011}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Alf\'e}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Cazorla}},\
and\ \bibinfo {author} {\bibfnamefont {M.~J.}\ \bibnamefont {Gillan}},\
}\bibfield {title} {\bibinfo {title} {The kinetics of homogeneous melting
beyond the limit of superheating},\ }\href
{https://doi.org/10.1063/1.3605601} {\bibfield {journal} {\bibinfo
{journal} {The Journal of Chemical Physics}\ }\textbf {\bibinfo {volume}
{135}},\ \bibinfo {pages} {024102} (\bibinfo {year} {2011})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Gonz\'alez-Cataldo}\ \emph
{et~al.}(2016)\citenamefont {Gonz\'alez-Cataldo}, \citenamefont {Davis},\
and\ \citenamefont {Gutierrez}}]{Gonzalez2016b}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {F.}~\bibnamefont
{Gonz\'alez-Cataldo}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Davis}},\ and\ \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Gutierrez}},\ }\bibfield {title} {\bibinfo {title} {{Z method calculations
to determine the melting curve of silica at high pressures}},\ }\href
{https://doi.org/10.1088/1742-6596/720/1/012032} {\bibfield {journal}
{\bibinfo {journal} {Journal of Physics: Conference Series}\ }\textbf
{\bibinfo {volume} {720}},\ \bibinfo {pages} {012032} (\bibinfo {year}
{2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Hernandez}\ \emph {et~al.}(2022)\citenamefont
{Hernandez}, \citenamefont {Mohn}, \citenamefont {Guren}, \citenamefont
{Baron},\ and\ \citenamefont {Tr{\o}nnes}}]{Hernandez2022}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.-A.}\ \bibnamefont
{Hernandez}}, \bibinfo {author} {\bibfnamefont {C.~E.}\ \bibnamefont {Mohn}},
\bibinfo {author} {\bibfnamefont {M.~G.}\ \bibnamefont {Guren}}, \bibinfo
{author} {\bibfnamefont {M.~A.}\ \bibnamefont {Baron}},\ and\ \bibinfo
{author} {\bibfnamefont {R.~G.}\ \bibnamefont {Tr{\o}nnes}},\ }\bibfield
{title} {\bibinfo {title} {Ab initio atomistic simulations of ca-perovskite
melting},\ }\href {https://doi.org/10.1029/2021GL097262} {\bibfield
{journal} {\bibinfo {journal} {Geophysical Research Letters}\ }\textbf
{\bibinfo {volume} {49}},\ \bibinfo {pages} {e2021GL097262} (\bibinfo {year}
{2022})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Di~Paola}\ and\ \citenamefont
{P.~Brodholt}(2016)}]{Brodholt2016}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.}~\bibnamefont
{Di~Paola}}\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{P.~Brodholt}},\ }\bibfield {title} {\bibinfo {title} {Modeling the melting
of multicomponent systems: the case of {MgSiO$_3$} perovskite under lower
mantle conditions},\ }\href {https://doi.org/10.1038/srep29830} {\bibfield
{journal} {\bibinfo {journal} {Scientific Reports}\ }\textbf {\bibinfo
{volume} {6}},\ \bibinfo {pages} {29830} (\bibinfo {year}
{2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Grabowski}\ \emph {et~al.}(2009)\citenamefont
{Grabowski}, \citenamefont {Ismer}, \citenamefont {Hickel},\ and\
\citenamefont {Neugebauer}}]{UP-TILD2009}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {B.}~\bibnamefont
{Grabowski}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Ismer}},
\bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Hickel}},\ and\ \bibinfo
{author} {\bibfnamefont {J.}~\bibnamefont {Neugebauer}},\ }\bibfield {title}
{\bibinfo {title} {Ab initio up to the melting point: Anharmonicity and
vacancies in aluminum},\ }\href {https://doi.org/10.1103/PhysRevB.79.134106}
{\bibfield {journal} {\bibinfo {journal} {Physical Review B}\ }\textbf
{\bibinfo {volume} {79}},\ \bibinfo {pages} {134106} (\bibinfo {year}
{2009})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Duff}\ \emph {et~al.}(2015)\citenamefont {Duff},
\citenamefont {Davey}, \citenamefont {Korbmacher}, \citenamefont {Glensk},
\citenamefont {Grabowski}, \citenamefont {Neugebauer},\ and\ \citenamefont
{Finnis}}]{TU-TILD2015}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~I.}\ \bibnamefont
{Duff}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Davey}},
\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Korbmacher}}, \bibinfo
{author} {\bibfnamefont {A.}~\bibnamefont {Glensk}}, \bibinfo {author}
{\bibfnamefont {B.}~\bibnamefont {Grabowski}}, \bibinfo {author}
{\bibfnamefont {J.}~\bibnamefont {Neugebauer}},\ and\ \bibinfo {author}
{\bibfnamefont {M.~W.}\ \bibnamefont {Finnis}},\ }\bibfield {title}
{\bibinfo {title} {Improved method of calculating ab initio high-temperature
thermodynamic properties with application to zrc},\ }\href
{https://doi.org/10.1103/PhysRevB.91.214311} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {91}},\ \bibinfo
{pages} {214311} (\bibinfo {year} {2015})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Lin}\ \emph {et~al.}(2003)\citenamefont {Lin},
\citenamefont {Blanco},\ and\ \citenamefont {Goddard}}]{2PT2003}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.-T.}\ \bibnamefont
{Lin}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Blanco}},\ and\
\bibinfo {author} {\bibfnamefont {I.}~\bibnamefont {Goddard}, \bibfnamefont
{William~A.}},\ }\bibfield {title} {\bibinfo {title} {The two-phase model
for calculating thermodynamic properties of liquids from molecular dynamics:
Validation for the phase diagram of lennard-jones fluids},\ }\href
{https://doi.org/10.1063/1.1624057} {\bibfield {journal} {\bibinfo
{journal} {The Journal of Chemical Physics}\ }\textbf {\bibinfo {volume}
{119}},\ \bibinfo {pages} {11792} (\bibinfo {year} {2003})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Zhu}\ \emph {et~al.}(2017)\citenamefont {Zhu},
\citenamefont {Grabowski},\ and\ \citenamefont {Neugebauer}}]{Melting2017}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {L.-F.}\ \bibnamefont
{Zhu}}, \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont {Grabowski}},\
and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Neugebauer}},\
}\bibfield {title} {\bibinfo {title} {Efficient approach to compute melting
properties fully from ab initio with application to cu},\ }\href
{https://doi.org/10.1103/PhysRevB.96.224202} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {96}},\ \bibinfo
{pages} {224202} (\bibinfo {year} {2017})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {de~Vita}(1992)}]{ADV1992}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{de~Vita}},\ }\bibfield {title} {\bibinfo {title} {The energetics of defaces
and impurities in metal and ionic materials from first principles},\
}\href@noop {} {\bibfield {journal} {\bibinfo {journal} {PhD Thesis}\ ,\
\bibinfo {pages} {Keele University}} (\bibinfo {year} {1992})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Wentzcovitch}\ \emph {et~al.}(1992)\citenamefont
{Wentzcovitch}, \citenamefont {Martins},\ and\ \citenamefont
{Allen}}]{Wentzcovitch1992}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {R.~M.}\ \bibnamefont
{Wentzcovitch}}, \bibinfo {author} {\bibfnamefont {J.~L.}\ \bibnamefont
{Martins}},\ and\ \bibinfo {author} {\bibfnamefont {P.~B.}\ \bibnamefont
{Allen}},\ }\bibfield {title} {\bibinfo {title} {Energy versus free-energy
conservation in first-principles molecular dynamics},\ }\href
{https://doi.org/10.1103/PhysRevB.45.11372} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {45}},\ \bibinfo
{pages} {11372} (\bibinfo {year} {1992})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Weinert}\ and\ \citenamefont
{Davenport}(1992)}]{Weinert1992}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {M.}~\bibnamefont
{Weinert}}\ and\ \bibinfo {author} {\bibfnamefont {J.~W.}\ \bibnamefont
{Davenport}},\ }\bibfield {title} {\bibinfo {title} {Fractional occupations
and density-functional energies and forces},\ }\href
{https://doi.org/10.1103/PhysRevB.45.13709} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {45}},\ \bibinfo
{pages} {13709} (\bibinfo {year} {1992})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Corgne}\ \emph {et~al.}(2005)\citenamefont {Corgne},
\citenamefont {Liebske}, \citenamefont {Wood}, \citenamefont {Rubie},\ and\
\citenamefont {Frost}}]{Corgne2005}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{Corgne}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Liebske}},
\bibinfo {author} {\bibfnamefont {B.~J.}\ \bibnamefont {Wood}}, \bibinfo
{author} {\bibfnamefont {D.~C.}\ \bibnamefont {Rubie}},\ and\ \bibinfo
{author} {\bibfnamefont {D.~J.}\ \bibnamefont {Frost}},\ }\bibfield {title}
{\bibinfo {title} {Silicate perovskite-melt partitioning of trace elements
and geochemical signature of a deep perovskitic reservoir},\ }\href
{https://doi.org/10.1016/j.gca.2004.06.041} {\bibfield {journal} {\bibinfo
{journal} {Geochimica et Cosmochimica Acta}\ }\textbf {\bibinfo {volume}
{69}},\ \bibinfo {pages} {485} (\bibinfo {year} {2005})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Tateno}\ \emph {et~al.}(2018)\citenamefont {Tateno},
\citenamefont {Hirose}, \citenamefont {Sakata}, \citenamefont {Yonemitsu},
\citenamefont {Ozawa}, \citenamefont {Hirata}, \citenamefont {Hirao},\ and\
\citenamefont {Ohishi}}]{Tateno2018}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Tateno}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Hirose}},
\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Sakata}}, \bibinfo
{author} {\bibfnamefont {K.}~\bibnamefont {Yonemitsu}}, \bibinfo {author}
{\bibfnamefont {H.}~\bibnamefont {Ozawa}}, \bibinfo {author} {\bibfnamefont
{T.}~\bibnamefont {Hirata}}, \bibinfo {author} {\bibfnamefont
{N.}~\bibnamefont {Hirao}},\ and\ \bibinfo {author} {\bibfnamefont
{Y.}~\bibnamefont {Ohishi}},\ }\bibfield {title} {\bibinfo {title} {Melting
phase relations and element partitioning in morb to lowermost mantle
conditions},\ }\href {https://doi.org/10.1029/2018JB015790} {\bibfield
{journal} {\bibinfo {journal} {Journal of Geophysical Research: Solid
Earth}\ }\textbf {\bibinfo {volume} {123}},\ \bibinfo {pages} {5515}
(\bibinfo {year} {2018})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Das}\ \emph {et~al.}(2020)\citenamefont {Das},
\citenamefont {Mohn}, \citenamefont {Brodholt},\ and\ \citenamefont
{Tr{\o}nnes}}]{Das2020}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {P.~K.}\ \bibnamefont
{Das}}, \bibinfo {author} {\bibfnamefont {C.~E.}\ \bibnamefont {Mohn}},
\bibinfo {author} {\bibfnamefont {J.~P.}\ \bibnamefont {Brodholt}},\ and\
\bibinfo {author} {\bibfnamefont {R.~G.}\ \bibnamefont {Tr{\o}nnes}},\
}\bibfield {title} {\bibinfo {title} {High-pressure silica phase
transitions: Implications for deep mantle dynamics and silica crystallization
in the protocore},\ }\href {https://doi.org/10.2138/am-2020-7299} {\bibfield
{journal} {\bibinfo {journal} {American Mineralogist}\ }\textbf {\bibinfo
{volume} {105}},\ \bibinfo {pages} {1014} (\bibinfo {year}
{2020})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Gonz\'{a}lez-Cataldo}\ \emph
{et~al.}(2016)\citenamefont {Gonz\'{a}lez-Cataldo}, \citenamefont {Davis},\
and\ \citenamefont {Gutierrez}}]{Gonzalez2016a}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {F.}~\bibnamefont
{Gonz\'{a}lez-Cataldo}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Davis}},\ and\ \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Gutierrez}},\ }\bibfield {title} {\bibinfo {title} {Melting curve of
{SiO$_2$} at multimegabar pressures: implications for gas giants and
super-earths},\ }\href {https://doi.org/10.1038/srep26537} {\bibfield
{journal} {\bibinfo {journal} {Scientific Reports}\ }\textbf {\bibinfo
{volume} {6}},\ \bibinfo {pages} {26537} (\bibinfo {year}
{2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Hirose}(2017)}]{Hirose2017}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {K.}~\bibnamefont
{Hirose}},\ }\bibfield {title} {\bibinfo {title} {Crystallization of silicon
dioxide and compositional evolution of the earth's core},\ }\href
{https://doi.org/10.1038/nature21367} {\bibfield {journal} {\bibinfo
{journal} {Nature}\ }\textbf {\bibinfo {volume} {543}},\ \bibinfo {pages}
{99} (\bibinfo {year} {2017})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Tr{\o}nnes}\ \emph {et~al.}(2019)\citenamefont
{Tr{\o}nnes}, \citenamefont {Baron}, \citenamefont {Eigenmann}, \citenamefont
{Guren}, \citenamefont {Heyn}, \citenamefont {L{\o}ken},\ and\ \citenamefont
{Mohn}}]{Tronnes2019}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {R.~G.}\ \bibnamefont
{Tr{\o}nnes}}, \bibinfo {author} {\bibfnamefont {M.~A.}\ \bibnamefont
{Baron}}, \bibinfo {author} {\bibfnamefont {K.~R.}\ \bibnamefont
{Eigenmann}}, \bibinfo {author} {\bibfnamefont {M.~G.}\ \bibnamefont
{Guren}}, \bibinfo {author} {\bibfnamefont {B.~H.}\ \bibnamefont {Heyn}},
\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {L{\o}ken}},\ and\
\bibinfo {author} {\bibfnamefont {C.~E.}\ \bibnamefont {Mohn}},\ }\bibfield
{title} {\bibinfo {title} {Core formation, mantle differentiation and
core-mantle interaction within earth and the terrestrial planets},\ }\href
{https://doi.org/10.1016/j.tecto.2018.10.021} {\bibfield {journal} {\bibinfo
{journal} {Tectonophysics}\ }\textbf {\bibinfo {volume} {760}},\ \bibinfo
{pages} {165} (\bibinfo {year} {2019})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Mermin}(1965)}]{Mermin1965}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {N.~D.}\ \bibnamefont
{Mermin}},\ }\bibfield {title} {\bibinfo {title} {Thermal properties of the
inhomogeneous electron gas},\ }\href
{https://doi.org/10.1103/PhysRev.137.A1441} {\bibfield {journal} {\bibinfo
{journal} {Physical Review}\ }\textbf {\bibinfo {volume} {137}},\ \bibinfo
{pages} {A1441} (\bibinfo {year} {1965})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Kresse}\ and\ \citenamefont {Hafner}(1993)}]{VASP93}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Kresse}}\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Hafner}},\ }\bibfield {title} {\bibinfo {title} {Ab initio molecular
dynamics for liquid metals},\ }\href@noop {} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {47}},\ \bibinfo
{pages} {558} (\bibinfo {year} {1993})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Kresse}\ and\ \citenamefont {Hafner}(1994)}]{VASP94}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Kresse}}\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Hafner}},\ }\bibfield {title} {\bibinfo {title} {Ab initio
molecular-dynamics simulation of the liquid-metal-amorphous-semiconductor
transition in germanium},\ }\href@noop {} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {49}},\ \bibinfo
{pages} {14251} (\bibinfo {year} {1994})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Bl\"ochl}(1994)}]{PAW94}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {P.~E.}\ \bibnamefont
{Bl\"ochl}},\ }\bibfield {title} {\bibinfo {title} {Projector augmented-wave
method},\ }\href@noop {} {\bibfield {journal} {\bibinfo {journal} {Physical
Review B}\ }\textbf {\bibinfo {volume} {50}},\ \bibinfo {pages} {17953}
(\bibinfo {year} {1994})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Kresse}\ and\ \citenamefont {Joubert}(1999)}]{PAW99}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Kresse}}\ and\ \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Joubert}},\ }\bibfield {title} {\bibinfo {title} {From ultrasoft
pseudopotentials to the projector augmented-wave method},\ }\href@noop {}
{\bibfield {journal} {\bibinfo {journal} {Physical Review B}\ }\textbf
{\bibinfo {volume} {59}},\ \bibinfo {pages} {1758} (\bibinfo {year}
{1999})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Perdew}\ \emph {et~al.}(1996)\citenamefont {Perdew},
\citenamefont {Burke},\ and\ \citenamefont {Ernzerhof}}]{PBE96}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.~P.}\ \bibnamefont
{Perdew}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Burke}},\ and\
\bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Ernzerhof}},\ }\bibfield
{title} {\bibinfo {title} {Generalized gradient approximation made simple},\
}\href@noop {} {\bibfield {journal} {\bibinfo {journal} {Physical Review
Letters}\ }\textbf {\bibinfo {volume} {77}},\ \bibinfo {pages} {3865}
(\bibinfo {year} {1996})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Armiento}\ and\ \citenamefont
{Mattsson}(2005)}]{Armiento2005}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {R.}~\bibnamefont
{Armiento}}\ and\ \bibinfo {author} {\bibfnamefont {A.~E.}\ \bibnamefont
{Mattsson}},\ }\bibfield {title} {\bibinfo {title} {Functional designed to
include surface effects in self-consistent density functional theory},\
}\href {https://doi.org/10.1103/PhysRevB.72.085108} {\bibfield {journal}
{\bibinfo {journal} {Phys. Rev. B}\ }\textbf {\bibinfo {volume} {72}},\
\bibinfo {pages} {085108} (\bibinfo {year} {2005})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Alf\`e}(2005)}]{Alfe2005}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Alf\`e}},\ }\bibfield {title} {\bibinfo {title} {Melting curve of {MgO}
from first-principles simulations},\ }\href
{https://doi.org/10.1103/PhysRevLett.94.235701} {\bibfield {journal}
{\bibinfo {journal} {Phys. Rev. Lett.}\ }\textbf {\bibinfo {volume} {94}},\
\bibinfo {pages} {235701} (\bibinfo {year} {2005})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Belonoshko}\ and\ \citenamefont
{Dubrovinsky}(1995)}]{Belonoshko1995}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~B.}\ \bibnamefont
{Belonoshko}}\ and\ \bibinfo {author} {\bibfnamefont {L.~S.}\ \bibnamefont
{Dubrovinsky}},\ }\bibfield {title} {\bibinfo {title} {Molecular dynamics of
stishovite melting},\ }\href {https://doi.org/10.1016/0016-7037(95)00071-7}
{\bibfield {journal} {\bibinfo {journal} {Geochimica et Cosmochimica Acta}\
}\textbf {\bibinfo {volume} {59}},\ \bibinfo {pages} {1883} (\bibinfo {year}
{1995})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Usui}\ and\ \citenamefont
{Tsuchiya}(2010)}]{Usui2010}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont
{Usui}}\ and\ \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont
{Tsuchiya}},\ }\bibfield {title} {\bibinfo {title} {Ab initio two-phase
molecular dynamics on the melting curve of {SiO$_2$}},\ }\href
{https://doi.org/10.1007/s12583-010-0126-9} {\bibfield {journal} {\bibinfo
{journal} {Journal of Earth Science}\ }\textbf {\bibinfo {volume} {21}},\
\bibinfo {pages} {801} (\bibinfo {year} {2010})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Millot}\ \emph {et~al.}(2015)\citenamefont {Millot},
\citenamefont {Dubrovinskaia}, \citenamefont {\u{C}ernok}, \citenamefont
{Blaha}, \citenamefont {Dubrovinsky}, \citenamefont {Braun}, \citenamefont
{Celliers}, \citenamefont {Collins}, \citenamefont {Eggert},\ and\
\citenamefont {Jeanloz}}]{Millot2015}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {M.}~\bibnamefont
{Millot}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont
{Dubrovinskaia}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{\u{C}ernok}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Blaha}},
\bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Dubrovinsky}}, \bibinfo
{author} {\bibfnamefont {D.~G.}\ \bibnamefont {Braun}}, \bibinfo {author}
{\bibfnamefont {P.~M.}\ \bibnamefont {Celliers}}, \bibinfo {author}
{\bibfnamefont {G.~W.}\ \bibnamefont {Collins}}, \bibinfo {author}
{\bibfnamefont {J.~H.}\ \bibnamefont {Eggert}},\ and\ \bibinfo {author}
{\bibfnamefont {R.}~\bibnamefont {Jeanloz}},\ }\bibfield {title} {\bibinfo
{title} {Shock compression of stishovite and melting of silica at planetary
interior conditions},\ }\href {https://doi.org/doi:10.1126/science.1261507}
{\bibfield {journal} {\bibinfo {journal} {Science}\ }\textbf {\bibinfo
{volume} {347}},\ \bibinfo {pages} {418} (\bibinfo {year}
{2015})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Andrault}\ \emph {et~al.}(2022)\citenamefont
{Andrault}, \citenamefont {Pison}, \citenamefont {Morard}, \citenamefont
{Garbarino}, \citenamefont {Mezouar}, \citenamefont {Bouhifd},\ and\
\citenamefont {Kawamoto}}]{Andrault2022}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Andrault}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Pison}},
\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Morard}}, \bibinfo
{author} {\bibfnamefont {G.}~\bibnamefont {Garbarino}}, \bibinfo {author}
{\bibfnamefont {M.}~\bibnamefont {Mezouar}}, \bibinfo {author} {\bibfnamefont
{M.~A.}\ \bibnamefont {Bouhifd}},\ and\ \bibinfo {author} {\bibfnamefont
{T.}~\bibnamefont {Kawamoto}},\ }\bibfield {title} {\bibinfo {title}
{Comment on: Melting behavior of {SiO$_2$} up to 120 {GPa} (andrault et al.
2020)},\ }\href {https://doi.org/10.1007/s00269-021-01174-2} {\bibfield
{journal} {\bibinfo {journal} {Physics and Chemistry of Minerals}\ }\textbf
{\bibinfo {volume} {49}},\ \bibinfo {pages} {3} (\bibinfo {year}
{2022})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Andrault}\ \emph {et~al.}(2020)\citenamefont
{Andrault}, \citenamefont {Morard}, \citenamefont {Garbarino}, \citenamefont
{Mezouar}, \citenamefont {Bouhifd},\ and\ \citenamefont
{Kawamoto}}]{Andrault2020}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Andrault}}, \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Morard}},
\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Garbarino}}, \bibinfo
{author} {\bibfnamefont {M.}~\bibnamefont {Mezouar}}, \bibinfo {author}
{\bibfnamefont {M.~A.}\ \bibnamefont {Bouhifd}},\ and\ \bibinfo {author}
{\bibfnamefont {T.}~\bibnamefont {Kawamoto}},\ }\bibfield {title} {\bibinfo
{title} {Melting behavior of {SiO$_2$} up to 120 {GPa}},\ }\href
{https://doi.org/10.1007/s00269-019-01077-3} {\bibfield {journal} {\bibinfo
{journal} {Physics and Chemistry of Minerals}\ }\textbf {\bibinfo {volume}
{47}},\ \bibinfo {pages} {10} (\bibinfo {year} {2020})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Hong}\ and\ \citenamefont {{van de
Walle}}(2016)}]{Hong2016}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Q.-J.}\ \bibnamefont
{Hong}}\ and\ \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {{van de
Walle}}},\ }\bibfield {title} {\bibinfo {title} {A user guide for sluschi:
Solid and liquid in ultra small coexistence with hovering interfaces},\
}\href {https://doi.org/10.1016/j.calphad.2015.12.003} {\bibfield {journal}
{\bibinfo {journal} {Calphad}\ }\textbf {\bibinfo {volume} {52}},\ \bibinfo
{pages} {88} (\bibinfo {year} {2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Shen}\ \emph {et~al.}(2016)\citenamefont {Shen},
\citenamefont {Jester}, \citenamefont {Qi},\ and\ \citenamefont
{Reed}}]{Shen2016}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont
{Shen}}, \bibinfo {author} {\bibfnamefont {S.~B.}\ \bibnamefont {Jester}},
\bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Qi}},\ and\ \bibinfo
{author} {\bibfnamefont {E.~J.}\ \bibnamefont {Reed}},\ }\bibfield {title}
{\bibinfo {title} {Nanosecond homogeneous nucleation and crystal growth in
shock-compressed {SiO$_2$}},\ }\href {https://doi.org/10.1038/nmat4447}
{\bibfield {journal} {\bibinfo {journal} {Nature Materials}\ }\textbf
{\bibinfo {volume} {15}},\ \bibinfo {pages} {60} (\bibinfo {year}
{2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Wang}\ \emph {et~al.}(2013)\citenamefont {Wang},
\citenamefont {Zhang}, \citenamefont {Liu},\ and\ \citenamefont
{Song}}]{ModifiedZ2013}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont
{Wang}}, \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Zhang}},
\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Liu}},\ and\ \bibinfo
{author} {\bibfnamefont {H.}~\bibnamefont {Song}},\ }\bibfield {title}
{\bibinfo {title} {Modified z method to calculate melting curve by molecular
dynamics},\ }\href {https://doi.org/10.1063/1.4798225} {\bibfield {journal}
{\bibinfo {journal} {The Journal of Chemical Physics}\ }\textbf {\bibinfo
{volume} {138}},\ \bibinfo {pages} {134101} (\bibinfo {year}
{2013})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Karavaev}\ \emph {et~al.}(2016)\citenamefont
{Karavaev}, \citenamefont {Dremov},\ and\ \citenamefont
{Pravishkina}}]{ClassicZ2016}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~V.}\ \bibnamefont
{Karavaev}}, \bibinfo {author} {\bibfnamefont {V.~V.}\ \bibnamefont
{Dremov}},\ and\ \bibinfo {author} {\bibfnamefont {T.~A.}\ \bibnamefont
{Pravishkina}},\ }\bibfield {title} {\bibinfo {title} {Precise calculation
of melting curves by molecular dynamics},\ }\href
{https://doi.org/10.1016/j.commatsci.2016.08.014} {\bibfield {journal}
{\bibinfo {journal} {Computational Materials Science}\ }\textbf {\bibinfo
{volume} {124}},\ \bibinfo {pages} {335} (\bibinfo {year}
{2016})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Mohn}\ \emph
{et~al.}(2009{\natexlab{a}})\citenamefont {Mohn}, \citenamefont {St{\o}len},\
and\ \citenamefont {Hull}}]{Mohn1CuI}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.~E.}\ \bibnamefont
{Mohn}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {St{\o}len}},\
and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Hull}},\ }\bibfield
{title} {\bibinfo {title} {Diffusion within ${\alpha}$\text{-CuI} studied
using ab initio molecular dynamics simulations},\ }\href
{https://doi.org/10.1088/0953-8984/21/33/335403} {\bibfield {journal}
{\bibinfo {journal} {Journal of Physics: Condensed Matter}\ }\textbf
{\bibinfo {volume} {21}},\ \bibinfo {pages} {335403} (\bibinfo {year}
{2009}{\natexlab{a}})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Mohn}\ \emph
{et~al.}(2009{\natexlab{b}})\citenamefont {Mohn}, \citenamefont {St{\o}len},
\citenamefont {Norberg},\ and\ \citenamefont {Hull}}]{MohnBi2O3}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.~E.}\ \bibnamefont
{Mohn}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {St{\o}len}},
\bibinfo {author} {\bibfnamefont {S.~T.}\ \bibnamefont {Norberg}},\ and\
\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Hull}},\ }\bibfield
{title} {\bibinfo {title} {Ab initio molecular dynamics simulations of
oxide-ion disorder in the ${\delta}$\text{-Bi$_2$O$_3$}},\ }\href
{https://doi.org/10.1103/PhysRevB.80.024205} {\bibfield {journal} {\bibinfo
{journal} {Phys. Rev. B}\ }\textbf {\bibinfo {volume} {80}},\ \bibinfo
{pages} {024205} (\bibinfo {year} {2009}{\natexlab{b}})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Alf\'{e}}\ \emph {et~al.}(2002)\citenamefont
{Alf\'{e}}, \citenamefont {Gillan},\ and\ \citenamefont {Price}}]{Alfe2002}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont
{Alf\'{e}}}, \bibinfo {author} {\bibfnamefont {M.~J.}\ \bibnamefont
{Gillan}},\ and\ \bibinfo {author} {\bibfnamefont {G.~D.}\ \bibnamefont
{Price}},\ }\bibfield {title} {\bibinfo {title} {Complementary approaches to
the ab initio calculation of melting properties},\ }\href
{https://doi.org/10.1063/1.1460865} {\bibfield {journal} {\bibinfo
{journal} {The Journal of Chemical Physics}\ }\textbf {\bibinfo {volume}
{116}},\ \bibinfo {pages} {6170} (\bibinfo {year} {2002})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Wilson}\ and\ \citenamefont
{Stixrude}(2021)}]{Wilson2021}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont
{Wilson}}\ and\ \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont
{Stixrude}},\ }\bibfield {title} {\bibinfo {title} {Entropy, dynamics, and
freezing of {CaSiO$_3$} liquid},\ }\href
{https://doi.org/10.1016/j.gca.2021.03.015} {\bibfield {journal} {\bibinfo
{journal} {Geochimica et Cosmochimica Acta}\ }\textbf {\bibinfo {volume}
{302}},\ \bibinfo {pages} {1} (\bibinfo {year} {2021})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Zhang}\ \emph {et~al.}(2022)\citenamefont {Zhang},
\citenamefont {Fan}, \citenamefont {Zhang}, \citenamefont {Li}, \citenamefont
{Wu},\ and\ \citenamefont {Li}}]{Zhang2022}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Zhang}}, \bibinfo {author} {\bibfnamefont {X.}~\bibnamefont {Fan}}, \bibinfo
{author} {\bibfnamefont {Q.}~\bibnamefont {Zhang}}, \bibinfo {author}
{\bibfnamefont {Q.}~\bibnamefont {Li}}, \bibinfo {author} {\bibfnamefont
{Y.}~\bibnamefont {Wu}},\ and\ \bibinfo {author} {\bibfnamefont
{M.}~\bibnamefont {Li}},\ }\bibfield {title} {\bibinfo {title} {Partial
disordering and homogeneous melting in multicomponent systems},\ }\href
{https://doi.org/10.1016/j.actamat.2022.118281} {\bibfield {journal}
{\bibinfo {journal} {Acta Materialia}\ }\textbf {\bibinfo {volume} {239}},\
\bibinfo {pages} {118281} (\bibinfo {year} {2022})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Belonoshko}\ \emph {et~al.}(2008)\citenamefont
{Belonoshko}, \citenamefont {Burakovsky}, \citenamefont {Chen}, \citenamefont
{Johansson}, \citenamefont {Mikhaylushkin}, \citenamefont {Preston},
\citenamefont {Simak},\ and\ \citenamefont {Swift}}]{Belonoshko2008}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~B.}\ \bibnamefont
{Belonoshko}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont
{Burakovsky}}, \bibinfo {author} {\bibfnamefont {S.~P.}\ \bibnamefont
{Chen}}, \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont {Johansson}},
\bibinfo {author} {\bibfnamefont {A.~S.}\ \bibnamefont {Mikhaylushkin}},
\bibinfo {author} {\bibfnamefont {D.~L.}\ \bibnamefont {Preston}}, \bibinfo
{author} {\bibfnamefont {S.~I.}\ \bibnamefont {Simak}},\ and\ \bibinfo
{author} {\bibfnamefont {D.~C.}\ \bibnamefont {Swift}},\ }\bibfield {title}
{\bibinfo {title} {Molybdenum at high pressure and temperature: Melting from
another solid phase},\ }\href
{https://doi.org/10.1103/PhysRevLett.100.135701} {\bibfield {journal}
{\bibinfo {journal} {Physical Review Letters}\ }\textbf {\bibinfo {volume}
{100}},\ \bibinfo {pages} {135701} (\bibinfo {year} {2008})}\BibitemShut
{NoStop}%
\bibitem [{\citenamefont {Rabuck}\ and\ \citenamefont
{Scuseria}(1999)}]{RN3736}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~D.}\ \bibnamefont
{Rabuck}}\ and\ \bibinfo {author} {\bibfnamefont {G.~E.}\ \bibnamefont
{Scuseria}},\ }\bibfield {title} {\bibinfo {title} {Improving
self-consistent field convergence by varying occupation numbers},\ }\href
{https://doi.org/10.1063/1.478177} {\bibfield {journal} {\bibinfo {journal}
{The Journal of Chemical Physics}\ }\textbf {\bibinfo {volume} {110}},\
\bibinfo {pages} {695} (\bibinfo {year} {1999})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Kresse}\ and\ \citenamefont
{Furthm\"uller}(1996{\natexlab{a}})}]{VASP96a}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Kresse}}\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Furthm\"uller}},\ }\bibfield {title} {\bibinfo {title} {Efficient iterative
schemes for ab initio total-energy calculations using a plane-wave basis
set},\ }\href@noop {} {\bibfield {journal} {\bibinfo {journal} {Physical
Review B}\ }\textbf {\bibinfo {volume} {54}},\ \bibinfo {pages} {11169}
(\bibinfo {year} {1996}{\natexlab{a}})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Kresse}\ and\ \citenamefont
{Furthm\"uller}(1996{\natexlab{b}})}]{VASP96b}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Kresse}}\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont
{Furthm\"uller}},\ }\bibfield {title} {\bibinfo {title} {Efficiency of
ab-initio total energy calculations for metals and semiconductors using a
plane-wave basis set},\ }\href {https://doi.org/10.1016/0927-0256(96)00008-0}
{\bibfield {journal} {\bibinfo {journal} {Computational Materials Science}\
}\textbf {\bibinfo {volume} {6}},\ \bibinfo {pages} {15} (\bibinfo {year}
{1996}{\natexlab{b}})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Levi}\ \emph {et~al.}(2020)\citenamefont {Levi},
\citenamefont {Ivanov},\ and\ \citenamefont {J\'onsson}}]{RN3745}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {G.}~\bibnamefont
{Levi}}, \bibinfo {author} {\bibfnamefont {A.~V.}\ \bibnamefont {Ivanov}},\
and\ \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {J\'onsson}},\
}\bibfield {title} {\bibinfo {title} {Variational density functional
calculations of excited states via direct optimization},\ }\href
{https://doi.org/10.1021/acs.jctc.0c00597} {\bibfield {journal} {\bibinfo
{journal} {Journal of Chemical Theory and Computation}\ }\textbf {\bibinfo
{volume} {16}},\ \bibinfo {pages} {6968} (\bibinfo {year}
{2020})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Pederson}\ and\ \citenamefont
{Jackson}(1991)}]{Pederson1991}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {M.~R.}\ \bibnamefont
{Pederson}}\ and\ \bibinfo {author} {\bibfnamefont {K.~A.}\ \bibnamefont
{Jackson}},\ }\bibfield {title} {\bibinfo {title} {Pseudoenergies for
simulations on metallic systems},\ }\href
{https://doi.org/10.1103/PhysRevB.43.7312} {\bibfield {journal} {\bibinfo
{journal} {Physical Review B}\ }\textbf {\bibinfo {volume} {43}},\ \bibinfo
{pages} {7312} (\bibinfo {year} {1991})}\BibitemShut {NoStop}%
\end{thebibliography}%