build_linked_cell_neighborhood_batched

torch_sim.transforms.build_linked_cell_neighborhood_batched(positions, cell, pbc, cutoff, n_atoms, self_interaction=False)[source]

Fully batched linked-cell neighbor list construction.

Drop-in replacement for build_linked_cell_neighborhood that processes all structures simultaneously using padded tensors, eliminating Python for-loops over structures. The algorithm mirrors the single-structure linked_cell but operates on all structures at once via padded 3D/4D tensors.

Parameters:

  • positions (Tensor) – Flat atomic positions (N_total, 3).

  • cell (Tensor) – Unit cell matrices, broadcastable to (n_structure, 3, 3).

  • pbc (Tensor) – PBC flags, broadcastable to (n_structure, 3).

  • cutoff (float) – Neighbor cutoff distance.

  • n_atoms (Tensor) – Number of atoms per structure (n_structure,).

  • self_interaction (bool) – Whether to include self-pairs.

Returns:

(mapping, system_mapping, cell_shifts_idx) — same format as build_linked_cell_neighborhood.

Return type:

tuple[Tensor, Tensor, Tensor]