Neural Network potential for Si by Gupta (2026) N=2 (in series of 3) v000

This is a general purpose potential for solid state Si. It uses Behler Parrinello symmetry functions as descriptors, has 7 layers (including the output layer), and uses tanh activation functions. The potential was trained on a modified GAP PRX Si dataset, where all the liquid Si configurations were removed to ensure better model performance for the solid state Si. The number `N2` indicates that this is the second potential in the series of 3 identically trained potentials with the same dataset but with different initial weights.