Kobe 2026¶
This page provides the instructions and tutorials for the session dedicated to Yambo.
Webpage of the event: ACM Second Asian School on High-Performance Computing and Artificial Intelligence.
General instructions for the tutorial session¶
We will run the calculations on the Fugaku supercomputer. The following instructions explain all the practical stuff that you need to know to run the calculations on Fugaku: read them before going to the tutorial page.
Get the files for the tutorial¶
The folder containing the tutorial files is a bit heavy, hence we suggest to copy it from a shared directory instead of downloading it from the internet as written at the beginning of the tutorial.
GW tutorial¶
Go to your user work directory /vol0300/data/hp250477/Students and copy the material from the shared folder:
$ cp /vol0003/mdt0/data/hp250477/Lecturers/u14341_Fulvio_Paleari/MoS2_HPC_GW_tutorial_Fugaku.tar.gz .
Once it is copied, extract the data using the tar command
$ tar -xvzf MoS2_HPC_GW_tutorial_Fugaku.tar.gz
BSE tutorial¶
Go to your user work directory /vol0300/data/hp250477/Students and copy the material from the shared folder:
$ cp /vol0003/mdt0/data/hp250477/Lecturers/u14341_Fulvio_Paleari/MoS2_HPC_BSE_tutorial_Fugaku.tar.gz .
Once it is copied, extract the data using the tar command
$ tar -xvzf MoS2_HPC_BSE_tutorial_Fugaku.tar.gz
Load Yambo¶
You need to add the following paths to your PATH environment variable in order to have the yambo executable:
fujitsu compilation¶
$ export PATH=$PATH:/vol0300/data/hp250477/Materials/Material_Science/bin_fj/
gcc compilation¶
export PATH=$PATH:/vol0300/data/hp250477/Lecturers/u10035_William_Dawson/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/vol0300/data/hp250477/Lecturers/u10035_William_Dawson/lib
Open an interactive session¶
In some cases you need to access a compute node interactively to run the executables. This is done with the following command:
$ pjsub --interact -L "node=1" -L "elapse=1:00:00" --sparam "wait-time=600" -g "hp250477" -S -x PJM_LLIO_GFSCACHE=/vol0003 "-L rscgrp=int"
If you want you can create a unix alias.
To run Yambo in interactive mode, use mpiexec before yambo
$ mpiexec -np <ntasks> yambo
Caution
Since this is specific for Fugaku, in the tutorials you will find only yambo, without mpiexec.
Don’t forget to add it!
Once you are done, you can close the interactive session simply typing
$ exit
Add a job to the queue¶
A submission script for Fugaku is provided as an example.
##!/bin/bash -l
#PJM -L 'node=1'
#PJM -N 'job_name'
#PJM --mpi 'max-proc-per-node=4'
#PJM -L 'rscgrp=small'
#PJM -L 'elapse=0:10:00'
#PJM -g hp250477
#PJM -x PJM_LLIO_GFSCACHE=/vol0003
# OMP threads export to exploit full node
export OMP_NUM_THREADS=$((48/(PJM_MPI_PROC/PJM_NODE)))
# load yambo
export PATH=$PATH:/vol0300/data/hp250477/Materials/Material_Science/bin_fj/
# run yambo
mpiexec yambo -F <some_input_file> -J <some_job_label> -C <some_out_label>
Change the script so that the allocated resources match the parallelization scheme you set in the input file.
Note: On Fugaku, it is most efficient to run Yambo on 4 MPI tasks maximum per node. If you want to include more tasks, change the number of nodes instead (do not go beyond 12 nodes max in this tutorial).
Save the lines above into a file named job.sh and submit the job typing
$ pjsub job.sh
The status of the jobs can be monitored via:
pjstat # to inspect the status of jobs
# (hint: make a unix alias, if you like)
pjdel <jobid> # to delete jobs in the queue
Visualization¶
We will want to generate images to visualize the result. To do so,
OpenOnDemand is the best solution. First go to the main dashboard. Then selectInteractive Apps and from there select Desktop. You can leave all options as the default and click Launch. This will bring you to a virtual desktop running on a Fugaku node. You can then open the terminal application from the bottom menu. Tools like gnuplot and python will be available from there for future steps in the tutorial.
Tutorials¶
Follow the tutorials listed below.