Skip to content

Latest commit

 

History

History
129 lines (114 loc) · 4.62 KB

README.md

File metadata and controls

129 lines (114 loc) · 4.62 KB
Raw

Strucscan

strucscan provides a lightweight Python-based framework for high-throughput material simulation that loops over a specified list of input structures and computes a specified list of properties on compute clusters with a queueing system or on the local host. The property calculations are represented as a pipeline of successive, interdependent steps which can easily be adapted and extended. The data is stored in a human-readable data tree with flat hierarchy. Strucscan performs a series of scalable and easily extendable pre-processing and post-processing steps and compiles the results in Python dictionaries for further evaluation. strucscan comes with interfaces to the VASP software package for ab-initio calculations. The VASP software itself is not included in this distribution.

Documentation

A detailed documentation can be found here.

Installation and setup

  1. clone repository into a <directory> of your choice. Please clone 'main' branch only.
  2. cd in your cloned strucscan directory and type
pip3 install .

  1. set-up ~/.strucscan resource file: copy .strucscan in your home directory and set it up according to your preferences.
    These configurations can be edited any time and are read in by strucsan at every start.
    Mandatory keys:

    • PROJECT_PATH: (str) top node of your data tree.
    • STRUCTURES_PATH: (str) top node of your structure pool.
    • RESOURCE_PATH: (str) path to configuration files for binaries, submission scripts, engines settings and further files (e.g. VASP pseudopotentials).

    Optional keys:

    • DEBUG: (bool) enables print commands for more insight. Default is False.
    • STRUCT_FILE_FORMAT: (str) structure file format of your structure files. Valid values are all formats comptabile with ase.io.read method. Default is cfg.
    • SLEEP_TIME: (int) Time in sec that strucscan will pause before starting the next monitoring loop. Default are 60 s.

Dependencies

  • ase
  • numpy
  • scipy
  • spglib

Resource directory

The resource directory contains script templates and configurations for modules and calls that can be tailored for specific machines. Additionally, you can deposit parameters and settings for the individual engines. The resource directory is organized like this:

resources
 ├── machineconfig
 │    ├── HPC1
 │    │    ├── config.yaml
 │    │    └── machinescripts
 │    │         ├── queue1.sge
 │    │         ├── queue2.sge
 │    │         └── ...
 │    │         
 │    ├── HPC2
 │    │    ├── config.yaml
 │    │    └── machinescripts
 │    │         ├── queue1.sge
 │    │         ├── queue2.sge
 │    │         └── ...
 │    │         
 │    └── ...
 │
 └── engines
      ├── vasp
      │    ├── bin
      │    ├── settings
      │    └── potentials
      │         ├── potpaw
      │         ├── potpaw_PBE
      │         └── potpaw_GGA
      │         
      ├── another_engine
      │    ├── bin
      │    ├── settings
      │    └── potentials
      │         
      └── ...

The machine configuration folder (machineconfig) contains the information that is required to start a serial or parallel calculation with the specific engine on the local host or to submit it do the scheduler of a compute cluster. This includes particularly modules that need to be loaded, the executable, and the queue requests in the config.yaml file as well as additional scripts that may be needed.

Example: machineconfig/example_vasp/config.yaml with parallel and serial executable of a VASP engine

VASP:
  parallel: | # this pipe is essential for reading multi-line entries
    module load vasp/mpi/5.4.4
    mpirun -np $NTOTALCORES vasp_std
  serial: | # this pipe is essential for reading multi-line entries
    module load vasp/serial/5.4.4
    vasp_std

Example: machineconfig/dummy/machinescripts/parallel12.sge with scheduler settings for parallel execution

#!/bin/bash 
#$ -S /bin/tcsh 
#$ -N [JOB_NAME] 
#$ -l qname=parallel12.q 
#$ -pe mpi12 [NTOTALCORES] 
#$ -e $JOB_ID.err 
#$ -o $JOB_ID.o 
#$ -cwd 
#$ -j y 
#$ -R y 
ipcrm --all 
#START=`date` 
#HOST=`hostname` 
#QNAME="parallel12" 
#echo "start: $START $HOSTNAME  $QNAME" > start.dat

Starting Strucscan

You can start strucscan from the command line using:

strucscan input.yaml

Several example calculations with input files are given in the notebooks in strucscan/examples.