Researchers use Deep Neural Network-based Quantum Simulations and Quasichemical Theory for Accurate Modeling of Molten Salt Thermodynamics

Molten salts are used in a wide range of applications because of their use as high-temperature heat-transfer mediums. Due to prospective applications in clean energy technologies like concentrated solar power storage and molten salt reactors, researchers have explored molten salt properties significantly in recent years. Due to their advantageous physicochemical characteristics (such as thermal conductivity, heat capacity, viscosity, etc.) and comparatively low manufacturing costs, molten salts make good candidates for these systems.

To find candidates with the best attributes, exploring a high-dimensional material space while building and optimizing salt combination compositions for different purposes is necessary. This increases the cost and difficulty of tests (such as X-ray and neutron diffraction and electrochemical measurements) in harsh environments.

Therefore, molten salt properties have been calculated using molecular dynamics (MD) simulations. To create accurate predictions, classical simulations effectively model systems over timescales of the order of nanoseconds.

Researchers at the University of Cincinnati, University of Massachusetts Lowell, and National Center for Computational Sciences, Oak Ridge National Laboratory collaborated for efficient and accurate modeling of solvation thermodynamics in molten salt liquids. For this, they computed the excess chemical potentials for the solute ions Na+ and Cl in the molten NaCl liquid using ab initio molecular dynamics (AIMD) simulations, and deep neural network interatomic potentials (NNIP), and quasichemical theory (QCT). This is the first time deep learning techniques have been used to calculate the excess thermodynamic properties of molten salts.

Compared to experiments, ab initio molecular dynamics (AIMD) simulations have captured local structure and solute chemistry more accurately. However, because AIMD requires expensive computation, it is challenging to access long time scales, and vast length scales to estimate the dynamic and thermal properties. The researchers demonstrate that using the NNIP model for molecular dynamics simulations significantly lowers the computation uncertainty to about 1 kcal mol1. These calculations are shown to be an effective method for directly computing the solute ion solvation free energy through molecular dynamics simulations. 

NNIP-MD approaches offer quantum-level accuracy with efficiency comparable to classical simulations. Therefore, the researchers believe they hold a significant amount of promise for their potential to play a significant part in exploring thermodynamic characteristics.

Quantitative studies of the activities, solubilities, and redox potentials of ionic species in the liquid phase are now possible thanks to the methodology, which paves the way for large-scale simulations of molten salt mixtures at a quantum mechanical level of accuracy.

The researchers believe this approach could offer crucial information for molten salt applications in concentrated solar energy storage systems and molten salt reactors.

This Article is written as a research summary article by Marktechpost Staff based on the research article 'Deep neural network based quantum simulations and quasichemical theory for accurate modeling of molten salt thermodynamics'. All Credit For This Research Goes To Researchers on This Project. Checkout the paper and reference article.

Please Don't Forget To Join Our ML Subreddit

Tanushree Shenwai is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Bhubaneswar. She is a Data Science enthusiast and has a keen interest in the scope of application of artificial intelligence in various fields. She is passionate about exploring the new advancements in technologies and their real-life application.

Jean-marc is a successful AI business executive .He leads and accelerates growth for AI powered solutions and started a computer vision company in 2006. He is a recognized speaker at AI conferences and has an MBA from Stanford.