GAP in CASTEP demo 2024/09/17

Getting started on Mahti

Either go to https://www.mahti.csc.fi/ and log in through the browser, or

start an interactive session

sinteractive --account project_2010950 --time 24:00:00 --cores 4

copy the demo files

cp -r /projappl/project_2010950/castep-ml/copy-this/ castep-ml-demo
cd castep-ml-demo

Further reading:

• documentation: https://libatoms.github.io/GAP/accelerated-aimd.html

Structure

code/ contains a copy of mace & hybrid-md tailored to our setup

examples/ contains a series of examples:

• 0-GAP-from-scratch GAP with TurboSOAP trained from scratch
• 1-MACE-only-checking MACE-MP0 with only checking accuracy
• 2-MACE-refit fine-tuning MACE-MP0 on-the-fly
• 3-MACE-custom-refit fine-tuning with user-supplied logic for refitting
• 4-try-your-own try this on your own system

Getting started:

load modules & paths

module use --append /projappl/project_2010950/modules
module load castep/cpu+QUIP+MACE

create a python environment

cd examples/
bash bootstrap.sh
source venv/bin/activate

or use /projappl/project_2010950/castep-ml/venv

Can run each