Virtual sites glitch through the box

[Koppeprojects]

The virtual particles displace themselves to the neighboring periodic box images. But the folded positions are still correct.

import numpy as np
import espressomd
import espressomd.virtual_sites as evs

system = espressomd.System(box_l=3 * [5])
system.time_step = 0.01
system.cell_system.skin = 0.4e-6
system.virtual_sites = evs.VirtualSitesRelative(have_quaternion=True)
center_part = system.part.add(
    pos=3 * [2.5],
    rotation=[False, True, False],
    fix=[True, True, True],
    gamma=1,
    gamma_rot=1
)

pos_virt = np.array(center_part.pos) + [1, 0, 0]
virtual_partcl = system.part.add(pos=pos_virt, director=[1, 0, 0], virtual=True)
virtual_partcl.vs_auto_relate_to(center_part)

system.thermostat.set_brownian(
    kT=1,
    gamma=1,
    seed=42,
    act_on_virtual=False,
)
system.integrator.set_brownian_dynamics()

positions = []
positionsfolded = []
for _ in range(100):
    system.integrator.run(100)
    positions.append(virtual_partcl.pos)
    positionsfolded.append(virtual_partcl.pos_folded)
distancefolded = np.linalg.norm(center_part.pos - np.array(positionsfolded), axis=1)
np.testing.assert_allclose(distancefolded, 1)

distance = np.linalg.norm(center_part.pos - np.array(positions), axis=1)
np.testing.assert_allclose(distance, 1)

[Koppeprojects]

For me I found a solution. I changed line 122 in the file VirtualSitesRelative.cpp from

fold_position(shift, image_shift, box_geo); 

to

fold_position(p.pos, image_shift, box_geo);

[jngrad]

Thank you for this bug report. As I understand it, this bug was introduced in 4.3-dev. A bugfix was proposed in #4564 and merged into the walberla branch, which in turn will be merged in the python branch in a few weeks.

[RudolfWeeber]

Simon, could you please check that this is indeed fixed in the Walberla branch?

[Koppeprojects]

It is fixed, the minimal script runs smoothly.