DP-GEN init
This tutorial is based on Bohrium and demonstrates how to perform DP-GEN init.
init has three subcommands, init_bulk, init_surf, and init_reaction
- init_bulk: Generating initial data for bulk systems.
- Relax in folder 00.place_ele
- Perturb and scale in folder 01.scale_pert
- Run a short AIMD in folder 02.md
- Collect data in folder 02.md.
- init_surf: Generating initial data for surface systems.
- Build specific surface in folder 00.place_ele
- Pertub and scale in folder 01.scale_pert
- init_reaction: Generating initial data for reactive systems.
- Exploring: LAMMPS
- Sampling: MDDatasetBuilder
- Labeling: Gaussian
For a detailed introduction and parameter description, please refer to DP-GEN's doc
Here, we take the Li-Si crystal structure as an example and explain the usage of init_bulk in detail.
Step 1: Prepare input files
Input file list
- INCAR files: INCAR_md INCAR_rlx
- POTCAR (one file per element)
- POSCAR (not necessarily needed)
- param.json
- machine.json
About INCAR
Two INCAR files are needed for init, namely INCAR_md and INCAR_rlx, corresponding to atomic relaxation and molecular dynamics calculation jobs, respectively.
In this example, the INCAR files are as follows:
INCAR_rlx
#Parameters
SYSTEM = dpgen_rlx
PREC = A
ISTART = 0
ICHARG = 2
#Electronic Relaxation
ENCUT = 650
NELM = 100
NELMIN = 6
NELMDL = -5
EDIFF = 1e-06
LREAL = .False.
ALGO = Fast # or Normal
#Ionic Relaxation
IBRION = 2
POTIM = 0.2
ISIF = 3
EDIFFG = -0.01
ISYM = 0
NSW = 100
ISMEAR = 0
SIGMA = 0.1
# Write flags
LWAVE = .False.
LCHARG = .False.
KSPACING = 0.16
KGAMMA = .False.
#KPAR = 1
#NPAR = 1
PSTRESS = 0.0
NCORE = 4
INCAR_md
#Parameters
SYSTEM = dpgen_md
PREC = A
ISTART = 0
ICHARG = 2
#Electronic Relaxation
ENCUT = 650
NELM = 100
NELMIN = 6
NELMDL = -5
EDIFF = 1e-06
LREAL = False
ALGO = Fast # or normal
#Ionic relaxation
IBRION = 0
ISIF = 2
#EDIFFG = -0.01 in
ISYM = 0
NSW = 10
ISMEAR = 0
SIGMA = 0.1
# MD related
SMASS = 0
POTIM = 2
TEBEG = 100
TEEND = 100
NBLOCK = 1
KBLOCK = 100
# Write flags
LWAVE = False
LCHARG = False
#parallel related
#KPAR = 4
#NPAR = 1
KSPACING = 0.16
KGAMMA = False
PSTRESS = 0.0
NCORE = 4
About param.json and machine.json
param.json and machine.json are the key parameter files for running dpgen. param.json tells dpgen what this calculation job is like, while machine.json tells dpgen what kind of environment you are in and what kind of machine you are using for the calculation.
This example param.json
{
"stages" : [1,2,3,4],
"elements": ["Li","Si"],
"cell_type": "diamond",
"latt": 10.0,
"super_cell": [1, 1, 1],
"potcars": ["..../POTCAR_Li","..../POTCAR_Si"],
"relax_incar": "./INCAR_rlx",
"md_incar" : "./INCAR_md",
"skip_relax": false,
"scale": [0.990,1.00,1.01],
"pert_numb": 10,
"pert_box": 0.03,
"pert_atom": 0.01,
"md_nstep" : 10,
"coll_ndata": 5000,
"type_map" : ["Li","Si"],
"_comment": "that's all"
}
Note:
- Check the path
Please check
"potcars","relax_incar", and"md_incar"to change the path of the user-specified file (absolute path or relative path). - Specify the conformation
If you want to use POSCAR for the init step, you need to add two parameters to specify the POSCAR file
"from_poscar": true,
"from_poscar_path": "<your-POSCAR>",
In the parameter file of this example, the "cell_type" and "latt" parameters are used to describe the conformation type and automatically generate a diamond-type LiSi alloy. DP-GEN version 0.10.6 and above supports "cell_type" values of "hcp", "fcc", "diamond", "sc" and "bcc". Please select the conformation according to your specific needs or the description in the literature.
- Specify the step Unlike dpgen run, which records steps with record.dpgen, init does not have this file. The way to specify the step is to check the current step in dpgen.log and use the "stages" parameter to specify the start and end steps. init_bulk has [1, 2, 3, 4] four steps, and init_surf has [1, 2] two steps. For example, in init_bulk, if the relaxation is successful but an error occurs during the perturbation in the second step, you can set "stages" to [2, 3, 4] and start from the second step.
machine.json:
{
"api_version": "1.0",
"deepmd_version": "2.1.0",
"train" :[
{
"command": "dp",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": ,
"keep_backup":true,
"input_data":{
"job_type": "indicate",
"log_file": "00*/train.log",
"grouped":true,
"job_name": "dpgen_train_job",
"disk_size": 100,
"scass_type":"c4_m15_1 * NVIDIA T4",
"checkpoint_files":["00*/checkpoint","00*/model.ckpt*"],
"checkpoint_time":30,
"platform": "ali",
"job_type": "container",
"image_address": "registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6" ,
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 4,
"gpu_per_node": 1,
"queue_name": "T4_4_15",
"group_size": 1
}
}
],
"model_devi":
[{
"command": "lmp -i input.lammps -v restart 0",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": ,
"keep_backup":true,
"input_data":{
"job_type": "indicate",
"log_file": "*/model_devi.log",
"grouped":true,
"job_name": "dpgen_model_devi_job",
"disk_size": 200,
"scass_type":"c4_m15_1 * NVIDIA T4",
"platform": "ali",
"job_type": "container",
"image_address": "registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6" ,
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 4,
"gpu_per_node": 1,
"queue_name": "T4_4_15",
"group_size": 50
}
}
],
"fp":
[{
"command": "limit -m unlimited; ulimit -s unlimited; mpirun vasp_std",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": ,
"input_data":{
"api_version":2,
"job_type": "indicate",
"log_file": "task*/fp.log",
"grouped":true,
"job_name": "dpgen_fp_job",
"disk_size": 100,
"scass_type":"c16_m32_cpu",
"platform": "ali",
"job_type": "container",
"image_address":"# Fill in the VASP image name here",
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 16,
"gpu_per_node": 0,
"queue_name": "CPU",
"group_size": 50,
"source_list": ["/opt/intel/oneapi/setvars.sh"]
}
}
]
}
Description:
- Images
In this example, we have specified the training part to use LBG_DeePMD-kit_2.0.1_v1 image, the exploration part to use LBG_DeePMD-kit_2.0.1_v1 image (which also includes the LAMMPS program combined with DeePMD-kit), and the labeling part to use the VASP image (if you need to use the VASP image, please provide your VASP authorization credentials to bohrium email).
If you want to change the computing image, you can specify it through the image_address in the machine.json file, such as if you want to use DeePMD-kit 2.2.1 version, just change the image name to registry.dp.tech/dptech/deepmd-kit:2.2.1-cuda11.6. The image name can be viewed on the "Image Center" page.
- Please pay attention to the "command" part of fp and the "source_list" under "resource". If there are errors in software call permissions, it may be related to these two parameters.
Preparation is complete, the folder directory is as follows:
$ ls
INCAR_md INCAR_rlx INCAR_scf POTCAR_Li machine.json param.json POTCAR_Si
Step 2: DP-GEN init
Run in the folder directory with the prepared files:
nohup dpgen init_bulk param.json machine.json 1>log 2>err&
Open Bohrium-Job Management and you can see the job that was just successfully submitted. init_bulk will do two fp steps, which are relaxation and molecular dynamics simulation.
After the job is completed, check the directory:
To see the directory structure, use the tree command, you need to apt install tree to install tree, and then use the tree command in the file directory to see the tree diagram.
.
├── dpdispatcher.log
├── dpgen.log
├── err
├── INCAR_md
├── INCAR_rlx
├── INCAR_scf
├── lisi.diamond.01x01x01
│ ├── 00.place_ele
│ │ ├── Backup files, etc
│ │ ├── INCAR
│ │ ├── POSCAR
│ │ ├── POSCAR.unit
│ │ ├── POTCAR
│ │ ├── run_1659588706.sh
│ │ ├── run_1659589787.sh
│ │ ├── run_1659590393.sh
│ │ ├── sys-0001-0001
│ │ │ ├── 806905fcd0a0ce5b7a072c7ce7e3c1dcfd497471_task_tag_finished
│ │ │ ├── CHG
│ │ │ ├── CHGCAR
│ │ │ ├── CONTCAR
│ │ │ ├── DOSCAR
│ │ │ ├── EIGENVAL
│ │ │ ├── fp.log
│ │ │ ├── IBZKPT
│ │ │ ├── INCAR -> ../INCAR
│ │ │ ├── OSZICAR
│ │ │ ├── OUTCAR
│ │ │ ├── PCDAT
│ │ │ ├── POSCAR
│ │ │ ├── POTCAR -> ../POTCAR
│ │ │ ├── REPORT
│ │ │ ├── vasprun.xml
│ │ │ ├── WAVECAR
│ │ │ └── XDATCAR
│ │ └── tmp_log
│ ├── 01.scale_pert
│ │ └── sys-0001-0001
│ │ └── scale-1.000
│ │ ├── 000000
│ │ │ └── POSCAR
│ │ ├── 000001
│ │ │ └── POSCAR
│ │ ├── 000002
│ │ │ └── POSCAR
│ │ ├── 000003
│ │ │ └── POSCAR
│ │ ├── 000004
│ │ │ └── POSCAR
│ │ └── POSCAR
│ ├── 02.md
│ │ ├── Backup files, etc
│ │ ├── INCAR
│ │ ├── POTCAR
│ │ ├── run_1659604259.sh
│ │ ├── sys-0001-0001
│ │ │ ├── deepmd
│ │ │ │ ├── box.raw
│ │ │ │ ├── coord.raw
│ │ │ │ ├── energy.raw
│ │ │ │ ├── force.raw
│ │ │ │ ├── set.000
│ │ │ │ │ ├── box.npy
│ │ │ │ │ ├── coord.npy
│ │ │ │ │ ├── energy.npy
│ │ │ │ │ ├── force.npy
│ │ │ │ │ └── virial.npy
│ │ │ │ ├── type_map.raw
│ │ │ │ ├── type.raw
│ │ │ │ └── virial.raw
│ │ │ └── scale-1.000
│ │ │ ├── 000000
│ │ │ │ ├── 00af13e72478f19ea730f67a982bb9c29dcd0c04_task_tag_finished
│ │ │ │ ├── CHG
│ │ │ │ ├── CHGCAR
│ │ │ │ ├── CONTCAR
│ │ │ │ ├── DOSCAR
│ │ │ │ ├── EIGENVAL
│ │ │ │ ├── fp.log
│ │ │ │ ├── IBZKPT
│ │ │ │ ├── INCAR -> ../../../INCAR
│ │ │ │ ├── OSZICAR
│ │ │ │ ├── OUTCAR
│ │ │ │ ├── PCDAT
│ │ │ │ ├── POSCAR
│ │ │ │ ├── POTCAR -> ../../../POTCAR
│ │ │ │ ├── REPORT
│ │ │ │ ├── vasprun.xml
│ │ │ │ ├── WAVECAR
│ │ │ │ └── XDATCAR
│ │ │ ├── 000001
│ │ │ │ └── ...
│ │ │ ├── 000002
│ │ │ │ └── ...
│ │ │ ├── 000003
│ │ │ │ └── ...
│ │ │ └── 000004
│ │ │ └── ...
│ │ └── tmp_log
│ └── param.json
├── log
├── machine.json
├── param.json
├── POTCAR_Li
└── POTCAR_Si
Step 3: Monitor Job Status and Results
Refer to Monitor Tasks for monitoring job status.
Refer to DP-GEN and Result Acquisition for obtaining execution results.
At this point, the dataset required for DP training is obtained through the DPGEN init_bulk process. The next step is to run DPGEN run for iterations and explore the sample space of the entire potential energy surface.
To monitor the job status, refer to Monitoring Tasks.
To obtain the results, refer to DP-GEN and Results Retrieval.
At this point, the autotest is completed, and you can view the results in the confs/mp-3034/<property-name> folder's result file. For example, in this case, the eos result can be found in confs/mp-3034/eos_00/result.out.