Supplementary Data: "Coherent and incoherent phonon transport in graphene/h-BN superlattice: a machine learning potential"

The graphene file contains a training set of isolated graphene structures, initially comprising 300 structures. Due to force errors arising from atomic distances, structures with distances greater than 20 Å were removed, resulting in a final training set of 269 structures.

The hBN file contains a training set of isolated h-BN, also originally consisting of 300 structures. Following the same criteria as for graphene, structures with force errors exceeding 20 Å were deleted, yielding a final training set of 281 structures.

The graphene-hBN file includes 300 isolated graphene structures, 300 isolated h-BN, and 320 graphene/h-BN interface structures. Similarly, structures with force errors above 20 Å were removed, resulting in a final training set of 851 structures.

All testing sets for potential function training are constructed in the same manner as their respective training sets.