[lammps-users] Atoms lost in parallel tempering

In parallel tempering, the dump file generated is missing some atom information
at certain timesteps.

You'll have to provide more info. What kind of
system. How many procs. How many replicas.
Is it reproducible. What do you mean: some atom
info is missing.

Steve

I have a system of 1700 atoms.

I am doing langevin fix at 4 different temperatures (4 processors).
#Fix

variable t world 0.90 1.0 1.1 1.2
fix tempfix all langevin $t $t 100.0 87678

#dump

dump 3 all atom 10000 dump.atom

#temper

temper 50000000 100000 $t tempfix 0 45466

I expect coordinates of 1700 atoms at every 10000 steps(dump parameter), but when I look at my dump file, I don’t find the coordinates of all the 1700 atoms at many timesteps. The information loss is reproducible. I am pasting a sample output for more clarity:

//Correct sample output

ITEM: TIMESTEP
0
ITEM: NUMBER OF ATOMS
1700
ITEM: BOX BOUNDS
-20 20
-20 20
-20 20
ITEM: ATOMS
1 1 0.5 0.543301 0.525

1700 4 0.5375 0.417802 0.705
ITEM: TIMESTEP
10000

//Incorrect Sample Output

ITEM: TIMESTEP
20000
ITEM: NUMBER OF ATOMS
1700
ITEM: BOX BOUNDS
-20 20
-20 20
-20 20
ITEM: ATOMS
1 1 0.114889 0.267638 0.500346

1696 4 0.111219 0.0915533 0.567453
1697 4 0.0621498 0.ITEM: TIMESTEP
30000
ITEM: NUMBER OF ATOMS
1700

One problem is that you are using the same dump file
for all replicas. Thus 4 different procs will be trying
to write to the same file. Try specifying dump.atom.$t and
see if you get 4 good files.

Steve

That works. Thank you.