IAP: Electronic-Structure Software Tutorials, 2016

Schedule: 2-5 pm, Room 8-119, Jan 12 and 14 (Octopus), Jan 19 and 21 (BerkeleyGW)

Instructor: Dr. David Strubbe, Department of Materials Science and Engineering, MIT

Please sign up by January 8 by emailing me at dstrubbe @ mit.edu, so we can be sure to have enough space and NERSC accounts.

It is important to bring your laptop so you can do the hands-on exercises.

overall intro slides

NERSC

For the BerkeleyGW part (and optionally the Octopus part) we will be running simulations on the Cori supercomputer at the National Energy Research Scientific Computing Center (NERSC). You will be provided with a training account to use the machine, which is strongly recommended even if you already have a NERSC account. If you are not familiar with NERSC, and especially if you are not familiar with other clusters or supercomputers, take a look over the information on NERSC's getting started guide. (Don't worry about the discussion of compiling and optimizing code though; the needed software has already been installed for you.)

To log in, run ssh trainX@cori.nersc.gov in your terminal, substituting the actual name of your training account.
Run your jobs from directories on the scratch disk, which has better I/O performance. cd $SCRATCH to get there.
Be aware that since this machine is far away, running X-Windows programs may be quite slow; better to do it locally.
You submit jobs to the SLURM scheduler by the sbatch command, e.g. sbatch job.scr, which will put them in the queue for execution when there is free space. Be sure you do this from the directory where your input files are located. The job script (see example below) specifies parameters about how many cores to use, what commands to run, etc.

You can see what jobs you currently have in the queue by executing squeue -u $USER, so you can see when your job finishes. A status code will be shown in the ST column: PD = pending, i.e. waiting in the queue, R = running, CG = completing. Example:

             JOBID       USER  ACCOUNT       NAME  PARTITION    QOS NODES   TIME_LIMIT       TIME   ST           START_TIME
            921487   dstrubbe    mp149 test_pulpo    regular normal     1      4:00:00       0:00   PD                  N/A

If you make a mistake, you can cancel a job by scancel JOBID, where JOBID is the number written by squeue.
After the run, you will have a file in the directory from which you submitted the job, called slurm-JOBID.out. If the job failed for some reason unrelated to the code, the error message will be there.
To copy files from cori to your local machine (e.g. to visualize or plot), in a terminal on your local machine, write scp trainX@cori.nersc.gov:FULL_PATH_TO_YOUR_FILE . (filling in the username and filename) and enter your password when prompted. You can get the full path by typing pwd in the terminal. For very small ASCII files, you may find cut and paste more convenient.

Sample job script using 4 cores of 1 node. Name it, for example, job.scr, and submit to run via sbatch job.scr.

#!/bin/bash
#SBATCH -J methane
#SBATCH -n 1
#SBATCH -p debug
#SBATCH -t 00:30:00
#SBATCH --export=ALL

cd $SLURM_SUBMIT_DIR
srun -n 4 /project/projectdirs/mp149/IAP2015/octopus-5.0.1/bin/octopus_mpi &> output

Octopus

Octopus: real-space TDDFT. tutorial materials

intro slides

Additional instructor: Jacob Sanders, Department of Chemistry and Chemical Biology, Harvard University

We are using version 5.0.1. The octopus tutorials can be run easily on a small machine such as your laptop or a cluster. For Mac OS, installation via MacPorts is recommended (see here). For Linux, install with a .deb file for your distribution, or along the lines of these instructions. Please be sure to run the testsuite with make check before trying any calculations.

BerkeleyGW

BerkeleyGW: many-body perturbation theory.

Additional instructor: Dr. Huashan Li, Department of Materials Science and Engineering, MIT

We are using version 1.2-beta. The tutorials are best run on the NERSC machines. However, if you are interested in installing on your own machine for later usage, you can use MacPorts for Mac OS (port install berkeleygw) for the last open release 1.1-beta2, or install from source following one of the files in the config directory of the source. Please be sure to run the testsuite with make check before trying any calculations.

Prof. Louie's introduction to the theory and applications: slides

Intro to practical issues for tutorial: slides

Bethe-Salpeter equation: slides

Instructions:

# Log in to Cori
ssh -X train##@cori.nersc.gov

# Load modules
module load berkeleygw/1.2-beta paratec

# Optional: make the output of 'ls' color-coded; only have to do this once
echo 'alias ls="ls --color"' >> ~/.bashrc.ext
. ~/.bashrc

# Start an interactive job.
# ** Do not start more than one interactive job! **
salloc -N 1 -p regular --reservation=20160121_tutorial -t 1:00:00

You will see a message like the following; answer yes.

The authenticity of host 'cmom06 (128.55.145.146)' can't be established.
RSA key fingerprint is SHA256:7SF90Sk0RfoVXDVqQfQtTbMx1/GW9wY0XSuEPnfCqRg.
Are you sure you want to continue connecting (yes/no)? yes

And then these messages:

salloc: Pending job allocation 969589
salloc: job 969589 queued and waiting for resources
salloc: job 969589 has been allocated resources
salloc: Granted job allocation 969589
salloc: Waiting for resource configuration
salloc: Nodes nid00182 are ready for job

When your interactive job has started, continue by doing

# Go to the scratch directory, where all runs should happen.
cd $SCRATCH

# List all examples available
ls /project/projectdirs/mp149/IAP2015

# Copy 1-silicon example to your directory
cp -R /project/projectdirs/mp149/IAP2015/1-silicon .

# Go to your local folder and follow instructions
cd 1-silicon
less README