[Code scan] VASP POSCAR negative scale factor is interpreted as a direct multiplier

[njzjz]

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

In VASP POSCAR files, a negative scale factor represents the target cell volume, not a direct negative multiplier for lattice vectors and Cartesian coordinates. The current parser multiplies the cell by scale unconditionally, so negative scales produce negative lattice vectors and the wrong volume.

Affected code:

dpdata/dpdata/formats/vasp/poscar.py

Lines 20 to 36 in a7a50bf

	scale = float(lines[1])
	cell = []
	move_flags = []
	for ii in range(2, 5):
	boxv = [float(jj) for jj in lines[ii].split()]
	boxv = np.array(boxv) * scale
	cell.append(boxv)
	system["cells"] = [np.array(cell)]
	natoms = sum(system["atom_numbs"])
	coord = []
	for ii in range(8, 8 + natoms):
	tmp = lines[ii].split()
	tmpv = [float(jj) for jj in tmp[:3]]
	if cartesian:
	tmpv = np.array(tmpv) * scale
	else:
	tmpv = np.matmul(np.array(tmpv), system["cells"][0])

Minimal reproducer:

import numpy as np
from dpdata.formats.vasp.poscar import to_system_data

lines = """test
-8
1 0 0
0 1 0
0 0 1
H
1
Direct
0 0 0
""".splitlines()

data = to_system_data(lines)
print(data["cells"][0])
print(np.linalg.det(data["cells"][0]))

Current behavior computes a cell with determinant -512. The expected cell volume is 8 for this input.

The parser should handle negative scale factors as target volumes, while preserving the current direct multiplier behavior for positive scale factors.

[dosubot]

The issue is confirmed. The current parser at lines 23-26 unconditionally multiplies lattice vectors by scale:

for ii in range(2, 5):
    boxv = [float(jj) for jj in lines[ii].split()]
    boxv = np.array(boxv) * scale
    cell.append(boxv)

Per VASP convention, when the scale factor is negative, its absolute value represents the target cell volume. The fix would be:

1. Read the raw (unscaled) lattice vectors
2. Compute the unscaled cell volume V₀ = |det(cell)|
3. Apply the actual scaling factor as (|scale| / V₀)^(1/3) to all lattice vectors (and Cartesian coordinates)

Something like:

scale = float(lines[1])
cell = []
for ii in range(2, 5):
    boxv = [float(jj) for jj in lines[ii].split()]
    cell.append(boxv)
cell = np.array(cell)

if scale < 0:
    # Negative scale means target volume
    vol_unscaled = abs(np.linalg.det(cell))
    scale = (abs(scale) / vol_unscaled) ** (1.0 / 3.0)

cell = cell * scale

For the reproducer's input (identity matrix, scale = -8), V₀ = 1, so the effective scale becomes 8^(1/3) = 2, giving a cell with volume 8 — matching the expected behavior.

There are currently no test cases covering negative scale factors [1], so a new test file with a negative scale POSCAR would be needed as well.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.