[Code scan] ABACUS MD can fabricate zero force labels when FORCE is absent

[njzjz]

This issue is part of a Codex global repository code scan.

The ABACUS MD dump reader initializes forces as a zero array for all frames and atoms, fills it only when the dump header contains FORCE, and later always assigns data["forces"] = force for LabeledSystem output. If an ABACUS MD dump contains positions and energies but no force columns, dpdata can return valid-looking zero forces instead of omitting forces or reporting that force labels are unavailable.

Affected code:

dpdata/dpdata/formats/abacus/md.py

Lines 57 to 75 in a7a50bf

	check_line = 10
	assert "POSITION" in dumplines[check_line], (
	"keywords 'POSITION' cannot be found in the 6th line. Please check."
	)
	if "FORCE" in dumplines[check_line]:
	calc_force = True
	nframes_dump = -1
	if calc_stress:
	nframes_dump = int(nlines / (total_natoms + 13))
	else:
	nframes_dump = int(nlines / (total_natoms + 9))
	assert nframes_dump > 0, (
	"Number of lines in MD_dump file = %d. Number of atoms = %d. The MD_dump file is incomplete." # noqa: UP031
	% (nlines, total_natoms)
	)
	cells = np.zeros([nframes_dump, 3, 3])
	stresses = np.zeros([nframes_dump, 3, 3])
	forces = np.zeros([nframes_dump, total_natoms, 3])

dpdata/dpdata/formats/abacus/md.py

Lines 105 to 124 in a7a50bf

	for iat in range(total_natoms):
	# INDEX LABEL POSITION (Angstrom) FORCE (eV/Angstrom) VELOCITY (Angstrom/fs)
	# 0 Sn 0.000000000000 0.000000000000 0.000000000000 -0.000000000000 -0.000000000001 -0.000000000001 0.001244557166 -0.000346684288 0.000768457739
	# 1 Sn 0.000000000000 3.102800034079 3.102800034079 -0.000186795145 -0.000453823768 -0.000453823768 0.000550996187 -0.000886442775 0.001579501983
	# for abacus version >= v3.1.4, the value of POSITION is the real cartessian position, and unit is angstrom, and if cal_force the VELOCITY is added at the end.
	# for abacus version < v3.1.4, the real position = POSITION * celldm
	coords[iframe, iat] = np.array(
	[float(i) for i in dumplines[iline + skipline + iat].split()[2:5]]
	)
	if not newversion:
	coords[iframe, iat] *= celldm
	if calc_force:
	forces[iframe, iat] = np.array(
	[
	float(i)
	for i in dumplines[iline + skipline + iat].split()[5:8]
	]
	)

dpdata/dpdata/formats/abacus/md.py

Lines 198 to 214 in a7a50bf

	for iframe in range(ndump):
	stress[iframe] *= np.linalg.det(cells[iframe, :, :].reshape([3, 3]))
	if np.sum(np.abs(stress[0])) < 1e-10:
	stress = None
	magmom, magforce = get_mag_force(outlines)
	data["cells"] = cells
	# for idx in range(ndump):
	# data['cells'][:, :, :] = cell
	data["coords"] = coords
	data["energies"] = energy
	data["forces"] = force
	data["virials"] = stress
	if not isinstance(data["virials"], np.ndarray):
	del data["virials"]
	data["orig"] = np.zeros(3)

Problematic flow:

• calc_force defaults to False.
• forces = np.zeros([nframes_dump, total_natoms, 3]) is allocated before parsing.
• The parser only overwrites forces[...] inside if calc_force:.
• The labeled data dict always receives data["forces"] = force.

Expected behavior: if the dump does not contain force columns, the parser should not create zero force labels. It should either omit forces when returning unlabeled-compatible data, or fail clearly when the caller requests a LabeledSystem that requires forces for this format.