Inconsistent output between two LAMMPS versions?

LAMMPS LAMMPS Mailing List Mirror
1

Hi there,

I stumbled on something surprising, I hope that you can clarify the issue..

By running my input scripts I found that they don't give the same
output on the versions:

LAMMPS (29 Sep 2011)
LAMMPS (27 Oct 2011)

I tried to strip them down the most that I could, and got:

######################## shear.in
shell rm nonaff.dat
include params.in
read_restart equilibrated
set group all image 0 0 0
change_box triclinic
minimize 1e-6 1e-6 100000 100000
write_restart sheared
variable i loop 100

label loop

  clear
  read_restart sheared
  fix def all deform 1 xy delta 0.001 units box
  run 1
  unfix def
  compute nonaffvalq all msd
  compute nonaff all displace/atom
  minimize 1e-6 1e-6 100000 100000
  thermo_style custom step pe xy xz yz lx c_nonaffvalq[4]
  run 0
  thermo_style custom step pe xy xz yz lx
  variable nonaffval equal sqrt(c_nonaffvalq[4])
  shell echo ${nonaffval} >> nonaff.dat
  uncompute nonaffvalq
  uncompute nonaff
  write_restart sheared
  next i

jump shear.in loop

2

I tried to strip them down the most that I could, and got:

I stripped it down more, and the issue is always there:

######################## shear.in
include params.in
shell rm nonaff.dat
read_restart equilibrated
set group all image 0 0 0
change_box triclinic
minimize 1e-6 1e-6 100000 100000
write_restart sheared
variable i loop 100
label loop
  clear
  read_restart sheared
  fix def all deform 1 xy delta 0.001 units box
  run 1
  unfix def
  compute nonaffvalq all msd
  minimize 1e-6 1e-6 100000 100000
  thermo_style custom step pe xy xz yz lx c_nonaffvalq[4]
  run 0
  thermo_style custom step pe xy xz yz lx
  variable nonaffval equal sqrt(c_nonaffvalq[4])
  shell echo ${nonaffval} >> nonaff.dat
  uncompute nonaffvalq
  write_restart sheared
  next i
jump shear.in loop

3

I don't know what this means. You're saying
the MSD is 10^8 larger in one case than
another. If they are not both epsilon, then
how is this possible.

Steve

4

Dear Steve,
thanks for your reply.
As far as I can see something funny is going on, but maybe I'm
mistaken, or problems show up only on my machine.

I attach a restart datafile ("equilibrated") so that anybody can run
my script starting from it, and the output (different!) that I get
with the two versions
LAMMPS (29 Sep 2011)
LAMMPS (27 Oct 2011)
in the file nonaff.dat (see shear.in for details).

The newer version seems to dump machine epsilons in that file, which
doesn't make much sense to me, as the particles move as a consequence
of the minimize line 16 in my script shear.in

Thanks a ton to anyone looking into this,

Davide

shear.in (722 Bytes)

equilibrated (77.2 KB)

nonaff-27-Oct-2011.dat (1.55 KB)

nonaff-29-Sept-2011.dat (1.53 KB)

5

I think the minimize bug I just posted a 8Nov patch
for was affecting this. My gues is the minimizer was
returning without doing anything if the uninit variable
happened to be set, so your incremental displacements
were 0.0.

So try it again with the patch and see if it's OK.

Steve

6

Dear Steve,

with the 8Nov patch the issue is solved.

Thanks (once again) a lot for your great work.

Best regards,

Davide