about the error: lost atoms

Dear all,
I am running an MD simulation with lammps .My simulation has two kinds of potential, lj/smooth/linear and harmonic/shift/cut.But the lj/smooth/linear potential is only effective when the distance between two atoms is less than equilibrium separation.
When I parallelly calculate this model using four cpu, there is no error at any time.However , it presents the error “Lost atoms: original 256767 current 220374” during the minimization, once I use more than four cpu. My program is written as follows.

dimension 2
boundary s s p
atom_style bond
units lj

read_data rou5.data

lattice hex 0.916486

pair_style lj/smooth/linear 0.8
pair_coeff * * 1 0.712718974 0.8
pair_coeff 3 3 1.25992 1 1.12246
pair_coeff 1 3 1.25992 1 1.12246
pair_coeff 2 3 1.25992 1 1.12246

bond_style harmonic/shift/cut
bond_coeff 1 0.60457 1.12246 1.9
bond_coeff 2 1.25992 1.12246 1.234706
bond_coeff 3 1.25992 1.12246 1.234706
bond_coeff 4 1.25992 1.12246 1.234706
communicate single cutoff 3.0 vel yes

neighbor 0.3 bin
neigh_modify every 10 delay 0 check no

region boundary block INF INF INF -0.1 INF INF
region solid block INF INF 138.3 INF INF INF
region plate block INF INF 0 138.2 INF INF

group boundary region boundary
group solid region solid
group plate region plate
group mobile subtract all boundary

fix 3 boundary setforce 0.0 0.0 0.0

minimize 1.0e-8 1.0e-10 5000 1000

compute new all temp
timestep 0.001
thermo 10000

reset_timestep 0

fix 1 all nve

velocity mobile create 0.001 887723 temp new
fix 2 solid smd cvel 2000 -0.0005 tether NULL 1000 NULL 0.0

compute stress_all all stress/atom virial
compute ske solid ke

variable Mises atom sqrt(1/2)sqrt((c_stress_all[1]-c_stress_all[2])^2+(c_stress_all[2]-c_stress_all[3])^2+(c_stress_all[3]-c_stress_all[1])^2+6((c_stress_all[4])^2+(c_stress_all[5])^2+(c_stress_all[6])^2))
variable v equal sqrt(2*c_ske/3058900)
variable f equal f_2[2]*256767

thermo_style custom step temp v_v v_f
thermo_modify temp new

dump 1 all custom 10000 frou5.dump.lammpstrj id type x y z c_stress_all[1] c_stress_all[2] c_stress_all[4] v_Mises
dump 2 all image 100000 all_stressx.*.jpg c_stress_all[1] type axes yes 0.5 0.01 size 1024 1024 zoom 0.7
dump_modify 2 amap min max cf 0.0 2 min blue max red boxcolor green backcolor white

run 10000000

May I have your suggestions ? Many thanks for your reply .

“Lost atom” has been discussed on the mailing list many many times. Please search the mailing list for its discussions.

Ray

Lost atoms are typically due to bad dynamics, which can be
for many reasons. See Section_errors.html for the error message
and ideas.

Steve

Dear all,
    I am running an MD simulation with lammps .My simulation has two kinds
of potential, lj/smooth/linear and harmonic/shift/cut.But the
lj/smooth/linear potential is only effective when the distance between two
atoms is less than equilibrium separation.
    When I parallelly calculate this model using four cpu, there is no
error at any time.However , it presents the error "Lost atoms: original
256767 current 220374" during the minimization, once I use more than four
cpu. My program is written as follows.

this error message has been discussed extremely frequently on the
mailing list, check the mailing list archives for previous discussions
and solutions.
there is not a single cause, but there are multiple options. from your
description, i would bet that your box boundaries in the data file are
not conforming to the requirements of using shrinkwrap boundary
conditions.

axel.