Estimation

First principle calculation of molten salt thermophysical properties refer to simulating small atom scale systems through DFT or other MD simulations. The size and length of the simulation depends on the property trying to be ascertained. For example, density and heat capacity only require systems in the hundreds of atoms whereas more complicated properties such as viscosity, diffusion or thermal conductivity require thousands and longer simulation times to converge.

image from doi: 10.1038/s42004-022-00684-6

The main drawback to any first principles simulated data is that the accuracy of data comes down to how well a given theoretical model captures the interactions between atoms, and the difficulty in scaling these calculations to more complicated and longer simulations. Great care needs to be taken to quantify the error generated by these calculations.

To advance the theory and characterize the interactions even more accurately, neural networks are currently being applied to molten salt systems to predict the inter-atomic potentials of a system. However, currently these networks are not generalizable, being heavily limited to the specific atomic system they are initially trained to. These neural networks are also reliant on “training” data sets to begin with still requiring experimental measurements. The goal however is for these networks to be completely generalizable, potentially over the entire periodic table.

Detailed information on the progression and use of computational methods in molten salts can be seen in these published articles:

https://www.nature.com/articles/s42004-022-00684-6

https://www.sciencedirect.com/science/article/abs/pii/S0022311522003221

https://www.sciencedirect.com/science/article/abs/pii/S0038092X20305193

Through Redlich-Kister (RK) expansion and Muggianu interpolation, density of higher-order molten salt systems have been successfully predicted.  Well characterized psuedo-binaries and pure components of the higher order systems are used to calculate binary interaction parameters through RK-expansion. These interaction parameters can then be used to predict the density of higher order systems through Muggianu interpolation of any composition, making it very attractive for filling in major gaps for difficult to produce compositions. This technique has been shown for predicting density in higher-order chloride and flouride molten salt systems:

https://www.sciencedirect.com/science/article/pii/S0009250921006515

https://www.sciencedirect.com/science/article/pii/S0009250922005383

The main drawback and difference between these types of techniques compared to first principles calculations is that these techniques absolutely require experimental data to begin with, whereas ab-initio is purely based on models. The production of pseudo-binary interaction terms requires the binary and pure salts of interest need to be very well characterized . The extrapolation to higher order systems still requires some experimental data for that higher order system to formulate higher-order interaction terms.  Finally, only density has been demonstrated to be predicted with this technique, however other properties have been theorized to be possible to estimate.

Redlich-Kister modeled density of the LiF-NaF-KF system at 1000 K. Units are in g/cc

Development of RK estimation techniques is a focus of the maintainers of MSTDB-TP, and RK psuedo-binary interaction terms have been tabulated and stored on the MSTDB-TP Gitlab project.