Data has been obtained to model safer and "cleaner" nuclear reactors.

The challenge lies in the fact that the industrially significant properties of liquid salts are difficult to measure experimentally due to high temperatures and corrosiveness. As a result, computational models similar to the one presented in the Journal of Molecular Liquids are essential for improving technologies for producing pure metals and in nuclear energy. This research is supported by a grant from the Russian Science Foundation.

Liquid salts represent a vast class of substances with a wide range of industrial applications. Material scientists are continuously enhancing the composition and characteristics of molten salt mixtures to increase the efficiency of producing titanium, calcium, aluminum, and other pure metals, as well as to advance the development of next-generation nuclear reactors.

Alongside solar and wind energy, nuclear power plants will play a crucial role in the gradual transition of the economy to a carbon-free model. Meanwhile, fusion reactors hold great promise but are still far from realization. At the same time, there exists another, more refined technology in the field of nuclear energy that specifically requires molten salts with carefully selected physical and chemical properties.

Salt-melt reactors (liquid salt reactors) are safer, more environmentally friendly, and more energy-efficient than those currently in use. Their implementation eliminates the risk of hydrogen explosions, as seen during the Fukushima nuclear power plant accident. In general, the operating pressure of such alternative reactors is close to atmospheric pressure (compared to 75–150 atmospheres in current nuclear plants), which is advantageous both in terms of safety and operating costs. Additionally, fuel can be loaded into liquid salt reactors without halting their operation. Moreover, the operating temperature in these reactors is approximately twice as high as in modern reactors. The higher the temperature, the greater the efficiency of electric and thermal energy production.

Furthermore, liquid salt reactors could potentially incinerate radioactive waste from currently prevalent nuclear reactors—such as neptunium-237, americium-237, and other so-called minor actinides. Currently, these hazardous wastes must be buried, which is extremely costly and does not effectively resolve the issue of their accumulation. For liquid salt reactors, these radioactive isotopes could serve as fuel.

To unlock the potential of liquid salts in nuclear energy and metallurgy, engineers need to understand the properties of these substances under various thermodynamic conditions. However, it is challenging for material scientists to provide this information due to the vast number of possible chemical compositions of molten salts. It is impossible to consider all combinations, especially through physical experiments, which are costly and labor-intensive due to the extremely high corrosiveness and temperatures of salt melts.

The study's first author, Nikita Rybin, a researcher at the Laboratory of Artificial Intelligence Methods for Material Development at Skoltech's Center for Artificial Intelligence, commented on the research results: “A computationally guided search for melts with specific physicochemical properties can significantly simplify and accelerate the development of next-generation nuclear reactors by minimizing the need for real experiments.

In this study, we presented and tested a methodology that allows for the calculation of thermophysical properties of salt melts at non-zero temperatures. The results of such calculations for a salt called FLiNaK (composition: LiF, NaF, KF) align with existing experimental data, so we will further examine salts of different compositions and analyze additional properties—this methodology will assist in material selection for next-generation reactors.”

The solution proposed by the scientific group for calculating the properties of liquid salts relies on machine-learned interatomic potentials and molecular dynamics simulations. The potentials are trained on results from calculations performed with quantum-mechanical precision. Without machine learning, such calculations would be computationally unmanageable.