Resending: About verlet/split

Hi Everybody

I had a question about using run_style verlet/split. To summarize, I am running a NPT simulation using a series of two fix npt (first is to equilibrate and the second for production). I sent this message earlier, but it does not seem to have appeared on the archives because it had a large attachment and needed approval.

The equilibration run works flawlessly for an indefinite number of time steps. When I start the production fix (the second fix npt), the simulation always fails at the 1st time step with a message - ERROR on abc-processor: Bond atoms missing on xyz- processor or similar. I don’t believe physics is a problem because the simulation can run using the first fix indefinitely. It appears that as the atoms migrate from one Kspace processor to another, the original run can continue alright but attempting to start a new run fails. If I run the first fix for just one time step, the simulation proceeds from the first fix to the second smoothly.

I first observed this problem in a larger and complex system but I am now using toluene as an example. I am running this system on 16 (4X2X2) processors for the real space and 2 processors (2X1X1) for the k space. I did not observe this error if I used just one k_space processor (i.e. 8 (2X2X2) processors for real space and 1 (1X1X1) processor for k_space). I have attached pasted the input script file and screen file for the k_space partition. (the log files do not show the error).

Maybe I missing something obvious, but I have been trying to figure this out for some time and I am going in circles. Any help would be appreciated.

Thanks

Ketan

---------------------------------Input script-------------------------------------

units real

neigh_modify delay 5 every 1

atom_style full

bond_style harmonic

angle_style harmonic

dihedral_style harmonic

pair_style lj/cut/coul/long/opt 9

pair_modify mix arithmetic

pair_modify tail yes

kspace_style pppm 1e-4

boundary p p p

processors * * * part 1 2 multiple

read_data data.Toluene600

replicate 2 1 1

timestep 1

run_style verlet/split

variable mass_g equal mass(all)/6.02e23

variable vol_cm3 equal vol*1.0e-24

variable Density equal v_mass_g/v_vol_cm3

variable SpecificVolume equal 1.0/v_Density

thermo_style custom step temp press vol etotal epair emol evdwl ecoul elong ebond eangle enthalpy pe ke v_Density v_SpecificVolume lx ly lz cpu

thermo 100

fix 1 all npt temp 300 300 100.0 iso 1.0 1.0 1000.0

run 4000 #fs

unfix 1

fix 1 all npt temp 300 300 100.0 iso 1.0 1.0 1000.0

run 1 #fs

unfix 1

write_restart restart.EndofRun

Can you try this in the most current version of LAMMPS
and without verlet/split. I.e. just run with verlet. I doubt
that verlet/split has anything to do with the issue.

Steve

Hi Steve

Thanks for the reply. I compiled the latest version of LAMMPS (17Feb12) with two different versions

1. with the replica and opt packages

With verlet/split, the simulation finishes the first npt and crashes at the beginning of the second fix. I have attached log.lammps.0-verletsplit and log.lammps.1-verletsplit as the log file for real and kspace.

2. without any packages
With verlet, the simulation proceeds from one fix npt to the next smoothly and finishes successfully. I have attached log.lammps-verlet as the log file for this run.

Could you please suggest anything else?

Thanks
Ketan

Note: I am using intel compiler so I had to add #include "lmptype.h" before #include "mpi.h" in balance.cpp

log.lammps-verlet (15.3 KB)

log.lammps.0-verletsplit (13.4 KB)

log.lammps.1-verletsplit (2.98 KB)

Can you post your 2 input scripts (fail, not fail) and
data file for a simple, short run (as small a system as possible).
Also, please try the verlet/split case w/out the opt pair style,
to see if that is part of the problem.

Steve

Hi Steve

I have attached two folders containing the input files, data files and other files (if they might help you diagnose the problem). This is a small atomistic system of about 10,800 atoms which can run on 16+2 cores using run_style verlet/split and 16 cores using run_style verlet/split . Yes, in accordance with your suggestion I did not use the opt styles of pairs.

Thanks for your help
Ketan

FAILVerletSplit.zip (540 KB)

NOFAILVerlet.zip (1.37 MB)

Hi Steve

I have attached the two folders containing the input files, data files and other files (if they might help you diagnose the problem). This is a small atomistic system of about 10,800 atoms which can be run on 16+2 cores using run_style verlet/split and 16 cores using run_style verlet .

The fact that no one else has reported any issues with verlet/split means that I am most probably missing something.

Thanks
Ketan

FAILVerletSplit.zip (540 KB)

NOFAILVerlet.zip (1.37 MB)

Found the problem - there will be a patch today.

Steve

Hey Steve

Thanks for looking into this issue and resolving it so quickly.

Ketan