First-principles molten salt phase diagrams through thermodynamic integration

Abstract: Precise prediction of phase diagrams in molecular dynamics (MD) simulations is challenging due to the simultaneous need for long time scales, large length scales and accurate interatomic potentials. We show that thermodynamic integration (TI) from low-cost force fields to neural network potentials (NNPs) trained using density-functional theory (DFT) enables rapid first-principles prediction of the solid-liquid phase boundary in the model salt NaCl. We use this technique to compare the accuracy of several DFT exchange-correlation functionals for predicting the NaCl phase boundary, and find that the inclusion of dispersion interactions is critical to obtain good agreement with experiment. Importantly, our approach introduces a method to predict solid-liquid phase boundaries for any material at an ab-initio level of accuracy, with the majority of the computational cost at the level of classical potentials.

• The paper introduces a novel thermodynamic integration method using a reversible pseudosupercritical pathway to predict accurate solid-liquid phase boundaries.
• The paper demonstrates that DFT functionals with dispersion corrections yield phase boundary predictions that align closely with experimental data.
• The paper leverages neural network potentials to balance computational cost and accuracy, paving the way for improved designs in reactors and material synthesis.

First-Principles Predictions of Molten Salt Phase Diagrams via Thermodynamic Integration

The paper, "First-principles molten salt phase diagrams through thermodynamic integration," presents a computational methodology aimed at accurately predicting the solid-liquid phase boundaries of molten salts using machine-learned potentials combined with thermodynamic integration (TI). The significance of this work lies in its potential application to nuclear reactors and thermal energy storage systems where molten salts serve as cooling agents or media for containing fuel and fission products. This study primarily utilizes the NaCl system as a benchmark to validate the proposed methodology.

The challenge in constructing accurate phase diagrams in molecular dynamics arises from the need to model long time and large length scales with accurate interatomic potentials. The study addresses this challenge by employing neural network potentials (NNPs), which are trained on density functional theory (DFT) data, and using these potentials within a TI framework. This approach efficiently estimates the solid-liquid phase boundary of NaCl with first-principles accuracy.

The study highlights several key findings:

1. Thermodynamic Integration Methodology: A pseudosupercritical pathway is employed for predicting phase equilibria, enabling a reversible thermodynamic transformation between the solid and liquid phases. The accuracy of NNPs overcomes the limitations of classical force fields typically used for such calculations.
2. Functional Sensitivity: A comparative evaluation of several DFT exchange-correlation (XC) functionals indicates that dispersion interactions significantly impact the accurate prediction of the NaCl solid-liquid phase boundary. Functionals that include dispersion interactions (e.g., PBE D2 and PBE D3) provide results that more closely align with experimental data across a wider range of pressures and temperatures.
3. Computational Efficiency: By leveraging NNPs, most computations can be efficiently conducted at the lower level of classical potentials, and the higher-cost NNPs are utilized strategically for critical evaluation points, thus balancing computational cost and accuracy.

The implications from this work could extend beyond the NaCl system, opening pathways for predicting phase diagrams of multi-component systems or assessing solubility limits, both of which are crucial in material science. Furthermore, the ability to predict phase transitions with NNPs trained to any first-principles data poses significant theoretical advancements in the modeling of condensed matter systems.

For future developments, integrating this methodology with more extensive materials databases and expanding its applicability towards complex chemical environments and multicomponent systems may enhance its utility within both academic research and practical engineering fields. Additionally, refining machine-learned potentials to capture non-local interactions more accurately could further improve the predictive power of this approach.

Such enhancements could facilitate faster and more accurate predictions in various industrial and scientific applications, optimizing processes like molten salt reactor design and high-temperature material synthesis. This research paves the way for more robust simulation frameworks in the field of chemistry and materials science, promoting efficient computational strategies to explore and design new materials.