Fix balance without updating the core list

LAMMPS LAMMPS General Discussion
,
1

Hello,

Is it possible to use fix balance to split the atoms equally across cores, and keep the list fixed without being updated, so each particle belong to core #1 will belong to the same core throughout the simulation?

In this system the particles dont have any interactions.

Thank you

Here is the script, if needed.
This is part of a bigger script, but this is a working example.

# -------- Time parameters--------------------

variable	maxsteps equal 2000000000
variable    totalsteps equal 50000000
variable    MD_time equal 8000
variable	mydt equal 1e-3 
variable	runsteps equal ${MD_time}/${mydt}
variable 	delta_t equal 1000

if "${runsteps} > ${maxsteps}" then  &
"variable    runsteps equal ${maxsteps}"

# -------- Brownian dynamics variables --------------------

variable    rng string gaussian
variable    seed equal 17145296
variable    temp equal 1.0
variable    gamma_t equal 1.0
variable    gamma_r equal 0.40839

# -------- Dimensions variables-----------

dimension   2

variable    x_max equal 100
variable    x_min equal  0
variable    y_max equal 30
variable    y_min equal  0
variable    N_par equal 50000

# -------- Particles ratio --------------------

variable    comp_ratio equal 1
variable    N_mag equal ceil(${N_par}*${comp_ratio})

units    lj

atom_style    hybrid dipole sphere

boundary    p f p
newton off
lattice none 1.0
atom_modify    map array sort 5000 0.0
region    box block ${x_min} ${x_max} ${y_min} ${y_max} -0.2 0.2
region    box_par block ${x_min} ${x_max} 5 25 -0.2 0.2
create_box    2 box
mass   * 1.0
timestep    ${mydt}
neighbor    0.4 bin
neigh_modify    every 1 delay 1 check yes

create_atoms 1 random ${N_mag} ${seed} box_par overlap 0.01

#comm_style tiled

# -------computes: MSD, RDF------------------

compute msd_all all msd 

# -------Thermo Fixes to print to file VORO, STRESS, CLUSTER------------------

compute areas all voronoi/atom neighbors yes
compute	peratom all stress/atom NULL virial

# -------Fixes------------------

fix step all brownian/sphere ${temp} ${seed} gamma_t ${gamma_t} gamma_r ${gamma_r} rng ${rng}

#fix bal all balance ${delta_t} 1.1 rcb

# -------Thermo Fixes to print to file MSD, RDF------------------

fix msd_avg_all all ave/time ${delta_t} 1 ${delta_t} c_msd_all[1] c_msd_all[2] c_msd_all[4] file msd_all.txt

fix xwalls all wall/reflect ylo 5 yhi 25

# -------Apply LJ pair style------------------

pair_style none

# -------- Dumping --------------------

dump    3 all custom ${delta_t} data.lin.xy.xyz id x y xu yu
dump_modify 3 append yes

# -------- Simulation main run --------------------
write_data data2.xyz

run ${runsteps}  upto
write_restart restart.in2d
2

No.

The assignment of individual atoms to sub-domains has little to do with the balance command, but is part of the domain decomposition parallelization. As atoms move around, they will be re-assigned to sub-domains before every update of the neighbor lists. The balance (or fix balance) command just shifts the sub-domain boundaries to optimize the distribution according to the imbalance weights.

If you don’t want the atoms to migrate between sub-domains, you must not use domain decomposition, i.e. only use one MPI process.

3

More importantly – why?

Anything that needs you to “pin” a particle to a particular process is going to be brittle and non-reproducible. Better to fix whatever it is, than artificially handicap a good performant domain decomposition system (which you can’t really do anyway without modifying the guts of the parallelization model).