Actinide Molten Salts: A Machine-Learning Potential Molecular Dynamics Study.

Area of Science:

• Nuclear materials science
• Computational chemistry
• Condensed matter physics

Background:

• Actinide molten salts are crucial for nuclear energy applications.
• Molecular-level understanding is vital for designing advanced nuclear technologies.
• Computational studies of actinides in molten salts are challenging due to complex electronic structures.

Purpose of the Study:

• To develop efficient computational protocols for studying actinide molten salts.
• To investigate the structure, thermodynamics, and dynamics of ThCl4-NaCl and UCl3-NaCl systems.
• To overcome limitations in simulating properties requiring extensive statistical sampling.

Main Methods:

• Employed a machine-learning approach to create accurate interatomic potentials.
• Utilized density functional theory (DFT) accuracy for the machine-learning potential.
• Performed long molecular dynamics (MD) simulations (nanoseconds) on large systems (10^3 atoms).

Main Results:

• Achieved significantly reduced computational cost for MD simulations.
• Obtained detailed information on bonding structures, thermodynamics, and dynamics across various temperatures.
• Observed notable changes in actinide coordination environments and coordination sphere lifetimes.

Conclusions:

• Machine learning potentials offer an efficient route to study complex actinide molten salt systems.
• The findings provide critical data for optimizing nuclear energy applications.
• Actinides in molten salts may deviate from established entropy-scaling laws.