########

Scripts in this directory can be used to generate the universal MD dataset, which can then be processed and employed as the training set for universal FlashMD training set.

`run_fine_ts_md.py` performs the MD routine as described in Appendix B.3 of the manuscript.

`rand_strucs_new_10000.xyz` is an ASE extxyz format file of the 10,000 structures randomly sampled from the PET-MAD dataset (details found in the PET-MAD preprint).

`main_job_X.sh` are slurm job files that can be used to submit the calculations to compute nodes with 4 GPUs, and use them in parallel to perform the MD simulations efficiently.

`create_jobs.sh` generates all the job scripts that together perfrom the simulation for all 10,000 structures.

`queue_jobs.sh` queues the generated job scripts to a slurm scheduler.



With everything properly set up (job scripts must be modified to the user's computing environment!), one can execute `create_jobs.sh`, then execute `queue_jobs.sh`. The resulting outputs and trajectories will be saved under `strXXX/` directories. We leave it up to the user to organize the ASE-trajectory files for data processing prior to FlashMD training.

########


The "process/" subfolder contains all the necessary scripts to process the MD trajectories
into a data format that can be used for training.
