ORCA workflow

Convert xyz files to orca input files by Python
Run orca calculation controlled by bash scripts
Extract information from orca by Python
As examples, I used part (12 configurations) of the QM7 dataset and calculated SPE with ORCA DLPNO-CCSD(T) method.

Notice

This is not built for slurm or other cluster managing systems.
ORCA ccsd(t) nproc should not exceed number of electrons
- This results in more problems with working on a cluster
I use the name thread, which refers a bash script that will start ORCA calculations in sequence.
- This might be confusing, and if you have a better name for it, please let me know.

Important Idea:

I persuaded myself that spliting tasks into threads cannot make full use of all resources. I will migrate to ready-built task managing system like celery, Task Spooler, and GNU parallel in the future. Personally and currently I prefer Task Spooler.

Chat with Copilot chatbot:

nqhq-lou:
Copilot, do you know about any code repo on github that could manage tasks like an HPC with slurm system on a PC? I have many tasks and I want to run them in a row, maximizing the usage of GPU and CPU, so I consider such a task managing code is important.

GitHub Copilot:
Yes, there are several open-source task management tools available on GitHub that can help you manage tasks on a PC with a slurm system. One such tool is "Task Spooler", which is a Unix batch system where the tasks are submitted to a queue and executed in order. Another tool is "GNU Parallel", which is a command-line tool that can execute multiple jobs in parallel. You can also consider using "Celery", which is a distributed task queue that can be used to distribute tasks across multiple machines.

file structure

orca_workflow
├── .env  # DO REMEMBER TO MODIFY THIS FILE
├── README.md
├── requirements.txt
├── inp  # ORCA input files
│   ├── [task_name].inp
│   └── ...
├── logs  # store log files and failed records
│   ├── failed
│   ├── [log_name].inp
│   └── ...
├── notebooks  # keep notebooks in this folder to make the project tidy
│   ├── gen_inp.ipynb
│   └── out2info.ipynb
├── out  # ORCA output files
│   ├── [task_name]  # this is a folder consists of all the output files of a single calculation
│   │   ├── [task_name].inp  # copied from `./inp` folder
│   │   ├── [task_name].out  # the output log file
│   │   └── ...
│   └── ...
├── scripts
│   ├── check_progress.sh  # check the currect finished / total tasks
│   ├── check_thread_duplicate.sh  # check if there are duplicate tasks names
│   ├── find_pids.sh  # find the pids of orca processes and bash control processes
│   ├── run_one_task.sh  # start a single calculation
│   ├── run_one_thread.sh  # start a single thread, which will start single calculations in sequence
│   ├── run_all_threads.sh  # start multiple threads in parallel
│   └── warn_run_at_PROJECT_ROOT
├── threads  # ORCA input files for threads, you could store thread files to othe folders
│   ├── [thread_name]
│   └── ...
├── xyz  # xyz files
│   ├── [task_name].xyz
│   └── ...
└── [other_files]

Also I provided the qm7_xyz.tar which consists of 7165 xyz files from QM7 dataset for you to try out the workflow.

To start with

Modify .env file
- Set PROJECT_ROOT and ORCA_PATH (the abs path)
Install requirements.txt
And follow the follow procedures

Procedures

Follow To start with section
Put xyzfiles in ./xyz folder
Generate .inp files and save to ./inp folder
- Using ./notebooks/gen_inp.ipynb
Split .inp files for threads and save to ./threads folder
- Using ./notebooks/gen_inp.ipynb
Run ORCA calculation controlled by bash scripts
- bash ./script/run_all_threads.sh [thread_dir]
  - Make sure that run the script at PROJECT_ROOT folder
- .out files will be placed in ./out folder
- Checkout ./logs/failed file and rerun the failed .inp files
  - based on the ./logs/failed file, create new thread dir and files and run run_all_threads.sh again
(If failed, really sorry to hear that)
- Just restart from run_all_threads script.
- The script could identify the finished tasks and skip them.
Extract information from .out files
- Using ./notebooks/extract.ipynb
- Outputs will be saved in ./notebooks folder

Out folder

There maybe working, finished, error empty files to indicated the current status
working indicates the calculation is not finished
finished indicates the calculation is finished
No such files, indicated should restart the calculation (error or interrupted or not started yet)

Why jobs failed?

ORCA error: bad symmetry applied, not reasonable calculation settings.
Hardware: not enough RAM.

More features required:

remove .env, use script_root=$(cd $(dirname "${BASH_SOURCE[0]}") && pwd) to load projroot
Parallelization in python
- Processing .inp files
- Extracting information from .out files
script/check_thread_duplicate.sh strategy: newline

Some tips

I tested acetaldehyde to find the best cal settings, and here is the result.

Here walltime = nproc * runtime

parameters

test on QM7 dataset
SCF affect on SPE: small. - TightSCF > VeryTightSCF > NormalSCF > LooseSCF - just use TightSCF
PNO has effect on SPE: - TightPNO > LoosePNO > NormalPNO - but still use NormalPNO
focus on TightSCF and Normal PNO - considering the runtime - DEF2-QZVPP is the best basis set for DLPNO-CCSD(T) calculations, for smaller SPE reached
restricted or unrestricted:
- use restricted, unrestricted takes 3 times more time

compare with other softwares

adopted scheme: TightSCF, NormalPNO
- runtime: 318.646 sec @ 24 nproc
- walltime: 7647.504 sec

	orbitals	scf_level	pno_level	extrapolate	runtime	walltime	HF_E	T_E	Corr_E	SPE	d_HF_E	d_T_E	d_Corr_E	d_SPE
54	DEF2-QZVPP	TightSCF	NormalPNO	False	318.646	7647.504	-152.983856	-0.025331	-0.646664	-153.630519	0.000000	0.000173	0.000165	0.000165
30	CC-PVQZ	TightSCF	NormalPNO	False	353.257	8478.168	-152.981670	-0.025358	-0.648127	-153.629796	0.002186	0.000146	-0.001298	0.000887
42	DEF2-TZVPP	TightSCF	NormalPNO	False	77.859	1868.616	-152.976644	-0.022820	-0.610751	-153.587395	0.007212	0.002684	0.036077	0.043288
7	CC-PVTZ	TightSCF	NormalPNO	False	119.836	2876.064	-152.970349	-0.023073	-0.612558	-153.582908	0.013507	0.002432	0.034270	0.047776
18	DEF2-TZVP	TightSCF	NormalPNO	False	62.229	1493.496	-152.975249	-0.022270	-0.601023	-153.576271	0.008607	0.003234	0.045806	0.054413

m062x ksdft
- restricted dft, default settings
- runtime: 108.782867 sec @ 24 nproc, def2qzvpp
  - ORCA DLPNO-CCSD(T) is 3 times the KSDFT runtime

	basis	hf_e_tot	ht_runtime	ksdft_e_tot	ksdft_runtime	mock_correlation_energy
0	ccpvtz	-152.970350	1.679003	-153.815120	68.692879	-0.844770
1	def2tzvp	-152.975248	0.527590	-153.817361	51.181625	-0.842112
2	ccpvqz	-152.981670	21.002635	-153.825548	147.125949	-0.843878
3	def2tzvpp	-152.976644	1.177629	-153.818895	84.558870	-0.842251
4	def2qzvpp	-152.983855	12.715615	-153.831484	108.782867	-0.847628

Gaussian
- %nproc=64
- %mem=200GB
- %chk=./chkfiles/QM7_00500.chk
- #p ccsd(T)/aug-cc-pvtz
- restricted cal
- walltime: 171.5 sec
- runtime: 10752.2 sec
- ORCA @ def2qzvpp
  - 71% the walltime, while orbital is 4 zeta in ORCA compared to 3 zeta in Gaussian
- for same orbital CC-PVTZ
  - walltime: ORCA 2876.064 sec, Gaussian 10752.2 sec
  - spe: ORCA -153.582908, Gaussian -153.5964995, Gaussian lower
  - spe difference: 0.0135915 Hartree, 8.53 kcal/mol
  - correlation energy: ORCA -0.612558, Gaussian -0.623726, Gaussian lower

{'nproc': 64,
 'mem': '200GB',
 'task': ['#p', 'single_point'],
 'method': 'ccsd(T)/aug-cc-pvtz',
 'charge': 0,
 'multiplicity': 1,
 'wall_time(sec)': 171.5,
 'cpu_time(sec)': 10752.2,
 'E_MP2': -153.5524044,
 'E_MP3': -153.5653544,
 'E_MP4D': -153.5794421,
 'E_MP4DQ': -153.5659729,
 'E_MP4SDQ': -153.5729361,
 'E_HF': -152.9727735,
 'E_CCSD': -153.5708637,
 'E_CCSD(T)': -153.5964995}

best settings inp file

! UHF DLPNO-CCSD(T) DEF2-QZVPP DEF2-QZVPP/C TightSCF NormalPNO UseSym
%PAL NPROC 24 END
%MAXCORE 3000
* xyz 0 1
C       0.99381998      0.00934000     -0.00221000
C       2.48917010      0.03821000     -0.01128000
O       3.13976991      1.04803003     -0.26596999
H       0.65238001     -1.01326003     -0.19271000
H       0.60306999      0.66173002     -0.78594001
H       0.62986001      0.33136999      0.97585001
H       2.98791993     -0.88891000      0.31445001
*

test on transient state of C2H6/ethan

optimized geometry

! HF DLPNO-CCSD(T) DEF2-QZVPP DEF2-QZVPP/C TightSCF NormalPNO UseSym
%PAL NPROC 18 END
%MAXCORE 2000
* xyz 0 1
C                  0.76185700    0.00002000    0.00006600
C                 -0.76188000    0.00001500   -0.00002300
H                 -1.15681000    0.06580600    1.01418600
H                 -1.15664300    0.84543600   -0.56416600
H                 -1.15663000   -0.91129100   -0.45012000
H                  1.15677300    0.91146100    0.44991600
H                  1.15660800   -0.84559300    0.56411800
H                  1.15683700   -0.06603300   -1.01419700
*

transient state

! HF DLPNO-CCSD(T) DEF2-QZVPP DEF2-QZVPP/C TightSCF NormalPNO UseSym
%PAL NPROC 18 END
%MAXCORE 2000
* xyz 0 1
C                  0.76838300    0.00004500    0.00002700
C                 -0.76834300    0.00000400    0.00001800
H                 -1.17120200   -0.15353700   -1.00035400
H                 -1.17121000   -0.78954500    0.63317400
H                 -1.17128100    0.94312200    0.36712400
H                  1.17105800   -0.78979400    0.63272600
H                  1.17100400   -0.15340200   -1.00029100
H                  1.17139100    0.94285800    0.36735300
*

DLPNO-CCSD(T) results

spe_C2H6: -79.698919476598
spe_C2H6_ts: -79.694505709045
spe_d: 0.004413767552989611

Gaussian CCSD(T) results

setting: '#p ccsd(T)/aug-cc-pvtz'
spe_C2H6: -79.67991
spe_C2H6_ts: -79.6754785
spe_d: 0.0044315

compare DLPNO-CCSD(T) with gaussian CCSD(T)
- 0.0000177 Hartree
- 0.0111273 kcal/mol
m062x results (by PySCF)

mol:
    atom: C2H6
    unit: ang
    basis: def2qzvpp
    max_memory: 8000
spe_C2H6: -79.81469773093517
spe_C2H6_ts: -79.81043365138069
spe_d: 0.00426407955447416

compare DLPNO-CCSD(T) with m062x
- 0.000149688 Hartree
- 0.09393 kcal/mol, less than 1 kcal/mol

THE END OF README

Name		Name	Last commit message	Last commit date
Latest commit History 23 Commits
.vscode		.vscode
logs		logs
notebooks		notebooks
out		out
scripts		scripts
threads		threads
xyz		xyz
.env		.env
.gitignore		.gitignore
README.md		README.md
log_fpath		log_fpath
qm7_xyz.tar		qm7_xyz.tar
requirements.txt		requirements.txt

nqhq-lou/ORCA_workflow

Folders and files

Latest commit

History

Repository files navigation

ORCA workflow

Notice

file structure

To start with

Procedures

Out folder

Why jobs failed?

More features required:

Some tips

parameters

compare with other softwares

test on transient state of C2H6/ethan

About

Topics

Resources

Stars

Watchers

Forks

Languages