Deriving rigid fragments#

Use molsysmt.topology.get_rigid_fragments() to partition a complete covalent graph at explicitly chosen source bond indices. This identifies connected fragments after those cuts; chemical rotatable-bond classification is provided separately by molsysmt.topology.get_rotatable_bonds().

import molsysmt as msm
import pandas as pd
from molsysmt.native import Topology

molsys = Topology(n_atoms=4)
molsys.bonds = pd.DataFrame({
    'atom1_index': [0, 1, 2], 'atom2_index': [1, 2, 3],
    'bond_type': ['covalent'] * 3})
molsys._chemical_states[0].connectivity_completeness = 'complete'
fragments = msm.topology.get_rigid_fragments(molsys, bond_indices=[1])
assert fragments['fragment_atom_indices'].tolist() == [0, 1, 2, 3]
assert fragments['fragment_offsets'].tolist() == [0, 2, 4]
assert fragments['atom_fragment_indices'].tolist() == [0, 0, 1, 1]
assert fragments['fragment_pairs'].tolist() == [[0, 1]]

The input may be any supported form carrying the required complete chemical connectivity, including standalone ChemicalStates and ChemicalStatesDict. A format that loses completeness metadata must supply it explicitly before this operation; the function does not assume that its listed bonds are all bonds. chemical_state selects one state, and structure_indices resolves state assignments when chemical_state='structure'. Coordinates are unused.

Memberships are packed int64 arrays with offsets. Source atom and bond indices are retained; fragment indices are assigned by their smallest source atom index. Disconnected components and isolated atoms become separate fragments. Repeated cuts are deduplicated. An empty atom domain gives offsets [0], empty memberships and branch arrays of shape (0, 2).

Every chosen bond must be covalent and a bridge in the original graph. Ring bonds fail even when several simultaneous ring cuts would disconnect the graph. Dative bonds, invalid indices and incomplete graphs also fail. No chemical state or coordinate array is changed. This tool does not choose a root, declare TORSDOF, or construct PDBQT records.