Discovering melting temperature prediction models of inorganic solids by combining supervised and unsupervised learning

The melting temperature is important for materials design because of its relationship with thermal stability, synthesis, and processing conditions. Current empirical and computational melting point estimation techniques are limited in scope, computational feasibility, or interpretability. We report the development of a machine learning methodology for predicting melting temperatures of binary ionic solid materials. We evaluated different machine-learning models trained on a dataset of the melting points of 476 non-metallic crystalline binary compounds using materials embeddings constructed from elemental properties and density-functional theory calculations as model inputs. A direct supervised-learning approach yields a mean absolute error of around 180 K but suffers from low interpretability. We find that the fidelity of predictions can further be improved by introducing an additional unsupervised-learning step that first classifies the materials before the melting-point regression. Not only does this two-step model exhibit improved accuracy, but the approach also provides a level of interpretability with insights into feature importance and different types of melting that depend on the specific atomic bonding inside a material. Motivated by this finding, we used a symbolic learning approach to find interpretable physical models for the melting temperature, which recovered the best-performing features from both prior models and provided additional interpretability.

Melting Is Well-Known, but Is It Also Well-Understood?

Contrary to continuous phase transitions, where renormalization group theory provides a general framework, for discontinuous phase transitions such a framework seems to be absent. Although the thermodynamics of the latter type of transitions is well-known and requires input from two phases, for melting a variety of one-phase theories and models based on solids has been proposed, as a generally accepted theory for liquids is (yet) missing. Each theory or model deals with a specific mechanism using typically one of the various defects (vacancies, interstitials, dislocations, interstitialcies) present in solids. Furthermore, recognizing that surfaces are often present, one distinguishes between mechanical or bulk melting and thermodynamic or surface-mediated melting. After providing the necessary preliminaries, we discuss both types of melting in relation to the various defects. Thereafter we deal with the effect of pressure on the melting process, followed by a discussion along the line of type of materials. Subsequently, some other aspects and approaches are dealt with. An attempt to put melting in perspective concludes this review.

Synthesis of Self-Assembled Single Atomic Layer Gold Crystals-Goldene

We report, for the first time, a technique to synthesize free-standing, one-atom thick 2D gold crystals (namely, goldene) and self-assembled 2D periodic arrays of goldene. High-resolution transmission electron microscopy (HRTEM) imaging of goldene revealed herringbone and honeycomb lattices, which are primarily gold surface features due to its reconstruction. Imaging of these surface-only features by a nonsurface characterization technique such as HRTEM is an unequivocal proof of the absence of three-dimensionality in goldene. Atomic force microscopy confirmed 1-2 Å thickness of goldene. High-resolution X-ray photoelectron spectroscopy (HR-XPS), selective area electron diffraction, and energy-dispersive X-ray spectroscopy confirmed the chemical identity of goldene. We discovered the phenomenon of electric field-induced self-assembly of goldene supracrystals with a herringbone structure and developed an electric field printing (e-print) technique for goldene arrays. Goldene showed a semiconductor response with a knee voltage of ∼3.2 V, and I/V spectroscopy revealed periodic room temperature Coulomb blockade oscillations. These observations are consistent with the theoretical calculations reported in the literature predicting enhanced Coulombic interactions between gold valence electrons and the nucleus in stable 2D gold. Goldene exhibited multiple, intense, and well-resolved optical absorption peaks and several fine bands across the UV-vis region, and we calculated its optical band gap to be 3.59 eV. Magnetic force microscopy measurements of goldene periodic arrays showed a ∼5 mV peak amplitude confirming its ferromagnetism. Optical and magnetic properties of goldene are consistent with those reported in the literature for 2D planar gold clusters with less than 12 atoms.

The phase diagrams of beryllium and magnesium oxide at megabar pressures

We performab initiosimulations of beryllium (Be) and magnesium oxide (MgO) at megabar pressures and compare their structural and thermodynamic properties. We make a detailed comparison of our two recently derived phase diagrams of Be (Wuet al2021Phys. Rev.B104014103) and MgO (Soubiran and Militzer 2020Phys. Rev. Lett.125175701) using the thermodynamic integration technique, as they exhibit striking similarities regarding their shape. We explore whether the Lindemann criterion can explain the melting temperatures of these materials through the calculation of the Debye temperature at high pressure. From our free energy calculations, we find that the melting line of both materials is well represented by the Simon-Glazel fitT_m(P) =T₀(1 +P/a)^1/c, whereT₀= 1564 K,a= 15.8037 GPa andc= 2.4154 for Be, whileT₀= 3010 K,a= 10.5797 GPa andc= 2.8683 for the MgO in the B1. For the B2 phase, we use the valuesa= 26.1163 GPa andc= 2.2426. Both materials exhibit negative Clapeyron slopes on the boundaries between the two solid phases that are strongly affected by anharmonic effects, which also influence the location of the solid-solid-liquid triple point. We find that the quasi-harmonic approximation underestimates the stability range of the low-pressure phases, namely hcp for Be and B1 for MgO. We also compute the phonon dispersion relations at low and high pressure for each of the phases of these materials, and also explore how the phonon density of states is modified by temperature. Finally, we derive secondary shock Hugoniot curves in addition to the principal Hugoniot curve for both materials, and study their offsets in pressure between solid and liquid branches.

A Review of the Melting Curves of Transition Metals at High Pressures Using Static Compression Techniques

The accurate determination of melting curves for transition metals is an intense topic within high pressure research, both because of the technical challenges included as well as the controversial data obtained from various experiments. This review presents the main static techniques that are used for melting studies, with a strong focus on the diamond anvil cell; it also explores the state of the art of melting detection methods and analyzes the major reasons for discrepancies in the determination of the melting curves of transition metals. The physics of the melting transition is also discussed.

Ab initio calculations of the elastic and thermodynamic properties of gold under pressure

The paper presents first-principles FP-LMTO calculations on the relative stability of fcc, bcc, hcp and dhcp gold under pressure. They were done in local density approximation (LDA), as well as in generalized gradient approximation (GGA) with and without spin-orbit interaction. Phonon spectra for the considered gold structures were obtained from LDA calculations within linear response theory and the contribution of lattice vibrations to the free energy of the system was determined in quasiharmonic approximation. Our thorough adjustment of FP-LMTO internal parameters (linearization and tail energies, the MT-sphere radius) helped us to obtain results that agree well with the available experimental phase relation Dubrovinsky et al (2007 Phys. Rev. Lett. 98 045503) between fcc and hcp structures of gold under pressure. The calculations suggest that gold compressed at room temperature successively undergoes the following structural changes: [Formula: see text]. The paper also presents the calculated elastic constants of fcc, bcc and hcp Au, the principal Hugoniot and the melting curve. Calculated results were used to construct the PT-diagram which describes the relative stability of the gold structures under study up to 500 GPa.

Melting in super-earths

We examine the possible extent of melting in rock-iron super-earths, focusing on those in the habitable zone. We consider the energetics of accretion and core formation, the timescale of cooling and its dependence on viscosity and partial melting, thermal regulation via the temperature dependence of viscosity, and the melting curves of rock and iron components at the ultra-high pressures characteristic of super-earths. We find that the efficiency of kinetic energy deposition during accretion increases with planetary mass; considering the likely role of giant impacts and core formation, we find that super-earths probably complete their accretionary phase in an entirely molten state. Considerations of thermal regulation lead us to propose model temperature profiles of super-earths that are controlled by silicate melting. We estimate melting curves of iron and rock components up to the extreme pressures characteristic of super-earth interiors based on existing experimental and ab initio results and scaling laws. We construct super-earth thermal models by solving the equations of mass conservation and hydrostatic equilibrium, together with equations of state of rock and iron components. We set the potential temperature at the core-mantle boundary and at the surface to the local silicate melting temperature. We find that ancient (∼4 Gyr) super-earths may be partially molten at the top and bottom of their mantles, and that mantle convection is sufficiently vigorous to sustain dynamo action over the whole range of super-earth masses.

Solid–liquid transitions of sodium chloride at high pressures

We investigate solid-liquid transitions in NaCl at high pressures using molecular dynamics simulations, including the melting curve and superheating/supercooling as well as solid-solid and liquid-liquid transitions. The first-order B1-B2 (NaCl-CsCl type) transition in solid is observed at high pressures besides continuous liquid structure transitions, which are largely analogous to the B1-B2 transition in solid. The equilibrium melting temperatures (T(m)) up to megabar pressure are obtained from the solid-liquid coexistence technique and the superheating-supercooling hysteresis method. Lindemann's vibrational and Born's mechanical instabilities are found upon melting. The Lindemann frequency is calculated from the vibrational density of states. The Lindemann parameter (fractional root-mean-squared displacement) increases with pressure and approaches a constant asymptotically, similar to the Lennard-Jones system. However, the Lindemann melting relation holds for both B1 and B2 phases to high accuracy as for the Lennard-Jonesium. The B1 and B2 NaCl solids can be superheated by 0.18T(m) and 0.24T(m), and the NaCl liquid, supercooled by 0.22T(m) and 0.32T(m), respectively, at heating or cooling rates of 1 K/s and 1 K/ps. The amount of maximum superheating or supercooling and its weak pressure dependence observed for NaCl are in accord with experiments on alkali halides and with simulations on the Lennard-Jones system and Al.

Self-diffusion in a liquid complex plasma

Self-diffusion has been experimentally studied in a two-dimensional underdamped liquid complex (dusty) plasma. It was found that the self-diffusion coefficient D increases linearly with the temperature T: D/omega(E)a2 = (0.019 +/- 0.007)(T/T(m) - 1), where T(m), omega(E), and a are the melting temperature, the Einstein frequency, and the mean particle separation, respectively. No superdiffusion was observed, whereas a subdiffusion occurred at temperatures close to melting.

Vibrational density of states and Lindemann melting law

We examine the Lindemann melting law at different pressures using the vibrational density of states (DOS), equilibrium melting curve, and Lindemann parameter delta(L) (fractional root-mean-squared displacement, rmsd, at equilibrium melting) calculated independently from molecular dynamics simulations of the Lennard-Jones system. The DOS is obtained using spectra analysis of atomic velocities and accounts for anharmonicity. The increase of delta(L) with pressure is non-negligible: delta(L) is about 0.116 and 0.145 at ambient and extreme pressures, respectively. If the component of rmsd normal to a reflecting plane as in the Debye-Waller-factor-type measurements using x rays is adopted for delta(L), these values are about 0.067 (+/-0.002) and 0.084 (+/-0.003), and are comparable with experimental and calculated values for face-centered-cubic elements. We find that the Lindemann relation holds accurately at ambient and high pressures. The non-negligible pressure dependence of delta(L) suggests that caution should be exerted in applying the Lindemann law to obtaining the high pressure melting curve anchored at ambient pressure.

Ultrahigh-Pressure Melting of Lead: A Multidisciplinary Study

Measurements of the melting temperature of lead, carried out to pressures of 1 megabar (10(11) pascal) and temperatures near 4000 kelvin by means of a laser-heated diamond cell, are in excellent agreement with the results of previous shock-wave experiments. The data are analyzed by means of first principles quantum mechanical calculations, and the agreement documents the reliability of current experimental and theoretical techniques for studies of melting at ultrahigh pressures. These studies have potentially wide-ranging applications, from planetary science to condensed matter physics.