User Tools

Site Tools


This is an old revision of the document!

Replica exchange of the disordering of a cluster

For this job we will use the cluster HYPATIA available at Empa. There we have access to parallel facilities with reserved nodes for the lecture. How to connect to HYPATIA:

Dear Student,

In order to be able to run simulations at high priority, today we will work on the Empa Cluster. We have created a personal account for you. Since the cluster is behind a firewall, we must connect to a gate machine (jumphost) to be allowed to access to the cluster. For security reasons, there are two temporary passwords that you should change to a personal password (can be the same for the gate and for the cluster.

Here the instructions to connect. Your username/password (EMPA-USER, TEMP-PASSW1, TEMP-PASSW2) are listed at the end of this message.

1) Decide a password (we will call it EMPA-PASSW )

2) connect to the jumphost:

ssh -X Password: TEMP-PASSW1

3) Accept the contract

4) Set a new password (input old password, TEMP-PASSW1, write new password, EMPA-PASSW)

5) Connect to hypatia: ssh -X hypatia password: TEMP-PASSW2

6) Accept the contract

7) Change your password as in point 4) using TMP-PASSW2 as old password and set EMPA-PASSW

</warning> ============================================

User-specific information (note: TMP-PASSW1 ist the password for jump1, that is the FIRST one, but is listed as second):


[you@hypatia ~]$ mmm-init
[you@hypatia ~]$ cd /mnt/scratch/YOURUSER/
[you@hypatia ~]$ cp -r /home/psd/exercise_7 .
[you@hypatia ~]$ cd exercise_7

The commands that you need to do to perform the exercise are, in this order:

[you@hypatia ~]$ qsub 00_run
[you@hypatia ~]$ ./01_adapt_files
[you@hypatia ~]$ ./02_reorder
[you@hypatia ~]$ ./03_extract_allaverages

===== Running the job =====

The script contains the directives for the queuing system, including 16 cores on one nodes reserved for the job.

#=== job name:
#PBS -N parallel 
#=== wall time limit (h:m:s)
#PBS -l walltime=1:00:00
#choice of the number of nodes and proc. per node
#PBS -l nodes=1:ppn=16 
#PBS -q short
#which queue
#=== memory usage
##PBS -l mem=1024mb
#=== join stdout and stderr
#PBS -j oe
# set environment variables
module unload mvapich2
module load openmpi
module load lammps/17Nov16/openmpi/2.0.1/gcc/4.9.4
cd $PBS_O_WORKDIR          
rm parallel.o* log.* screen*
mpiexec -np 16 lmp_mpi -partition 16x1   -in input

The last line is the command to run a parallel lammps job with the input file input

===== The input file for lammps =====

The file input contains information for the program lammps. Details on the documentation can be found here

There is an initialization section, showing the kind of units (see this page), the dimensionality, the boundary conditions.

# Initialization
units            metal
dimension        3
boundary         p p p
atom_style       atomic

In the second part of the input file a spherical region is defined (to confine the cluster). Then the atoms are read from input.dat. We also assign a mass to the kind number 1 (there is just one atomic type for Argon).

region rs sphere 0 0 0 12.66
read_data        input.dat
mass 1 39.948

Then, we define the parameters for the Lennard-Jones potential. The units are eV for epsilon, and angstrom for sigma. The last number is the cutoff, in Angstrom.

pair_style lj/cut 8.5
pair_coeff 1 1 0.01042 3.405  8.5

Then, we initialize the fix and the velocity as well as the temperature of each replica, which have been previously generated using the program t.x present in the same directory. We distribute the temperature exponentially between 2 K and 40 K. In LAMMPS, a fix is any operation that is applied to the system during timestepping or minimization. Here we have a fix for controlling temperature with NVT (a different temperature for each temperature), and a fix for applying a harmonic restraint to the spherical region confining the cluster. In this way, the atoms going beyond this region will be elastically pushed back into the sphere.

variable i equal part
variable t world  2.00    2.44    2.98    3.64    4.45    5.43    6.63    8.09    9.88   12.07   14.74   17.99   21.97   26.83   32.76   40.00
velocity        all        create        $t        293288
velocity all zero linear
velocity all zero angular
fix        1        all        nvt        temp        $t        $t        0.1
fix  2 all wall/region rs harmonic 2.0 0.0 0.4

The next section is about writing out each 1000 steps the relecant information about temperature and energy. We also dump a restart file at the end, and every 10000 steps a structure in xyz format.

thermo         1000
thermo_style        custom        step  temp     pe ke etotal
thermo_modify       line        one
restart 5000000 restart.*
dump         2 all xyz 10000 structure_$
dump_modify  2 element "Ar" sort id

Finally, this is the command to run the tempering, with an exchange move attempted every 1000 step of molecular dynamics and an initial temperature $t that is different from replica to replica. The last numbers are random seeds that are used for choosing which replica to exchange and for the Metropolis criterion.

temper 5000000 1000 $t 1 3678 3490

===== Adapting the output files =====

We must now make some postprocessing on the output files. The goal is to performs averages at different temperatures. These averages are enhanced by the exchanges that were performed between different molecular dynamics replica. Note that temperature is set by a thermostat.

Example. Processor 0 starts with temperature T0=2 K, processor 1 with temperature T1=2.44 K. After 1000 steps, an exchange step is attempted and accepted with some probability (see theory slides, and also the paper doi::10.1063/1.481671
exercises/2017_ethz_mmm/replica_2017.1494429048.txt.gz · Last modified: 2017/05/10 15:10 by dpasserone