Is PBC=[True, True, True] for all MPTrj datasets?

[RylieWeaver]

Hi everyone! I’ve been working with the MPTrj dataset and noticed that each sample includes a "lattice_mat" but no explicit "pbc" field. My workflow reads the lattice and multiplies by fractional coordinates to get atomic positions, like this:

lattice_mat = torch.tensor(info["atoms"]["lattice_mat"], dtype=torch.float32)
coords = torch.tensor(info["atoms"]["coords"], dtype=torch.float32)

# Multiply to get Cartesian positions
positions = torch.matmul(lattice_mat, coords.T).T

However, there’s no "pbc" entry that I can find in the JSON (unlike other data sources I've used). I'm wondering if I should assume that pbc = [True, True, True] for all structures since it seems each structure is a fully periodic 3D crystal cell.

Questions:
(1) Should we always assume pbc = [True, True, True] for all MPTrj entries, given that we have a complete lattice_mat and fractional coordinates?
(2) Is there any case in MPTrj (like surfaces or molecules) that isn’t fully periodic?

Thanks in advance!

@bowen-bd

[bowen-bd]

Hi there! For all MPtrj containing datapoints, pbc should be [True, True, True], even for the slabs.

[RylieWeaver]

Got it. Thanks for the clarification!