Computational methods to simulate molten salt thermophysical properties

Abstract

Molten salts are important thermal conductors used in molten salt reactors and solar applications. To use molten salts safely, accurate knowledge of their thermophysical properties is necessary. However, it is experimentally challenging to measure these properties and a comprehensive evaluation of the full chemical space is unfeasible. Computational methods provide an alternative route to access these properties. Here, we summarize the developments in methods over the last 70 years and cluster them into three relevant eras. We review the main advances and limitations of each era and conclude with an optimistic perspective for the next decade, which will likely be dominated by emerging machine learning techniques. This article is aimed to help researchers in peripheral scientific domains understand the current challenges of molten salt simulation and identify opportunities to contribute.

TrIP─Transformer Interatomic Potential Predicts Realistic Energy Surface Using Physical Bias

“…This issue was recently exemplified in a thorough review of machine learning potentials used for Molten Salt simulations. 34 Here, it was revealed that fixed-length input vectors with implicit atom-type encoding within established deep learning-based molecular dynamic packages introduced a systematic error, rendering the accurate simulation of intricate salt systems problematic. This underscores the necessity for users to grasp the intricacies of the machine learning models they employ, ensuring a comprehensive comprehension of their boundaries and careful interpretation of their results.…”

“…A lack of understanding models’ limitations and the nuances of their training process can lead to erroneous conclusions drawn from simulations. This issue was recently exemplified in a thorough review of machine learning potentials used for Molten Salt simulations . Here, it was revealed that fixed-length input vectors with implicit atom-type encoding within established deep learning-based molecular dynamic packages introduced a systematic error, rendering the accurate simulation of intricate salt systems problematic.…”

Insights into the local structure evolution and thermophysical properties of NaCl–KCl–MgCl2–LaCl3 melt driven by machine learning

“…As a result, there is a growing need for precise descriptions of molten salts using computational methods. 13 Over the course of the past seven decades, several computational schemes have been devised for molten salt systems. These schemes encompass classical molecular dynamics (CMD), ab initio molecular dynamics (AIMD), and molecular dynamics with machine learning methods.…”

“…25 This advancement has ushered molecular dynamics simulation into a new era. 13 Machine learning molecular dynamics is a computational approach that utilizes data from DFT to model the potential energy surface (PES) of research systems. Over the past decade, significant advancements and notable achievements have been obtained by the use of machine learning molecular dynamics.…”

FY2022 Progress Report on Viscosity and Thermal Conductivity Measurements of Molten Salts

“…This effort is more multifaceted than simply measuring pure, binary, or higher-order salt mixtures that have not yet been measured. This effort also involves (1) establishing measurement techniques that offer repeatable, accurate, and validatable results for better characterized salts so that they can be then applied to the more poorly characterized salts; (2) identifying salts or salt mixtures that have been measured for some thermophysical properties but have large discrepancies from study to study, or high intrinsic experimental uncertainty, and targeting these salts for measurements in addition to entirely uncharacterized salts; (3) comparing the results of experimental measurements with theoretical or ab initio modeling techniques for modeling validation [8,9]; and (4) using binary subsystem data to estimate thermophysical properties with Redlich-Kister expansions, which involves extracting binary interaction parameters from fitting to experimental data and using Muggianu interpolation to model higher-order systems [10,11].…”