add DPRc implementation

examples/run/dprc/README.md

@@ -0,0 +1,5 @@
+## Deep Potential Range Correction (DPRc)
+
+This example shows how to train a DPRc model.
+
+For details, refer to [10.1021/acs.jctc.1c00201](https://doi.org/10.1021/acs.jctc.1c00201).

examples/run/dprc/generator.yaml

@@ -0,0 +1,215 @@
+type_map:
+  - C
+  - H
+  - O
+  - "N"
+  - P
+  - HW
+  - OW
+init_data_sys:
+  - /path/to/initial_data
+# qm_region: the Amber mask of the QM region in the parm file
+qm_region:
+  - ":1"
+# qm_charge: the charge of the QM region
+qm_charge:
+  - 0
+# parm7: absolute path to parm7 file
+parm7:
+  - /path/to/MON.parm7
+# mdin: absolute path to mdin file
+mdin:
+  - /path/to/ml.mdin
+# disang: absolute path to disang template file
+disang:
+  - /path/to/mon.disang
+# sys_configs: abosulute path to initial configurations
+sys_configs:
+  - - /path/to/init_-1.20.rst7
+    - /path/to/init_-0.80.rst7
+    - /path/to/init_-0.40.rst7
+    - /path/to/init_0.00.rst7
+# r: reaction coordinates to replace RVAL in the template; can be a list for 2D
+r:
+  - - -1.20
+    - -0.80
+    - -0.40
+    - 0.00
+numb_models: 4
+# same as the simplify
+default_training_param:
+  model:
+    # type_map: same as above
+    type_map:
+      - C
+      - H
+      - O
+      - "N"
+      - P
+      - HW
+      - OW
+    descriptor:
+      type: hybrid
+      list:
+        # QM-QM descriptor
+        - type: se_a
+          # sel: refer https://docs.deepmodeling.org/projects/deepmd/en/latest/model/sel.html to obatin the proper sel
+          # In a DPRc model, sel of MM atoms for the QM-QM descriptor should be set to zero.
+          sel:
+            - 6
+            - 7
+            - 6
+            - 1
+            - 1
+            - 0
+            - 0
+          rcut_smth: 1.0
+          rcut: 9.0
+          neuron:
+            - 25
+            - 50
+            - 100
+          resnet_dt: false
+          axis_neuron: 12
+          # exclude_types: In the QM-QM descriptor, MM-MM and QM-MM descriptor should be disabled.
+          # >>> from itertools import combinations_with_replacement, product
+          # >>> qm = range(5)
+          # >>> mm = range(5,7)
+          # >>> list(map(list,list(combinations_with_replacement(mm,2))+list(product(qm,mm))))
+          exclude_types: [[5, 5], [5, 6], [6, 6], [0, 5], [0, 6], [1, 5], [1, 6], [2, 5], [2, 6], [3, 5], [3, 6], [4, 5], [4, 6]]
+          # set_davg_zero: In a DPRc model, it must be set to true.
+          set_davg_zero: true
+          type_one_side: true
+          precision: "float32"
+        # QM-MM descriptor
+        - type: se_a
+          # *1.2 for MM atoms
+          sel:
+            - 6
+            - 7
+            - 6
+            - 1
+            - 1
+            - 88
+            - 44
+          rcut_smth: 1.0
+          rcut: 6.0
+          neuron:
+            - 25
+            - 50
+            - 100
+          resnet_dt: false
+          axis_neuron: 12
+          # exclude_types: In the QM-MM descriptor, QM-QM and MM-MM descriptor should be disabled.
+          # >>> from itertools import combinations_with_replacement, product
+          # >>> qm = range(5)
+          # >>> mm = range(5,7)
+          # >>> list(map(list,list(combinations_with_replacement(qm,2))+list(combinations_with_replacement(mm,2))))
+          exclude_types: [[0, 0], [0, 1], [0, 2], [0, 3], [0, 4], [1, 1], [1, 2], [1, 3], [1, 4], [2, 2], [2, 3], [2, 4], [3, 3], [3, 4], [4, 4], [5, 5], [5, 6], [6, 6]]
+          set_davg_zero: true
+          type_one_side: true
+          precision: "float32"
+    fitting_net:
+      neuron:
+        - 240
+        - 240
+        - 240
+      resnet_dt: true
+      # atom_ener: In a DPRc model, the atom_ener of MM atoms should be set to 0
+      atom_ener:
+        - null
+        - null
+        - null
+        - null
+        - null
+        - 0.0
+        - 0.0
+      precision: "float32"
+  # learning_rate: from 1.e-3 to 5.e-8
+  learning_rate:
+    type: exp
+    start_lr: 1.e-3
+    decay_steps: 400
+    stop_lr: 5.e-8
+  loss:
+    start_pref_e: 0.02
+    limit_pref_e: 1
+    start_pref_f: 1000
+    limit_pref_f: 1
+    start_pref_v: 0
+    limit_pref_v: 0
+  training:
+    numb_steps: 400000
+    disp_file: lcurve.out
+    disp_freq: 100
+    save_freq: 1000
+    disp_training: true
+    time_training: true
+    profiling: false
+    profiling_file: timeline.json
+dp_compress: true
+
+model_devi_f_trust_lo: 0.08
+model_devi_f_trust_hi: 0.25
+training_reuse_iter: 2
+
+
+# nsteps: the number of steps to run
+nsteps:
+- 10000
+model_devi_jobs:
+  - sys_idx: [0] 
+    trj_freq: 40
+  #1 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #2 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #3
+  - sys_idx: [0] 
+    trj_freq: 40
+  #4 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #5 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #6 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #7 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #8 
+  - sys_idx: [0] 
+    trj_freq: 40
+  #9
+  - sys_idx: [0] 
+    trj_freq: 40
+  #10
+  - sys_idx: [0] 
+    trj_freq: 40
+
+# fp_task_max: the maximum fp tasks to calculate
+fp_task_max: 1000
+fp_task_min: 10
+fp_params:
+  low_level_mdin: /path/to/low_level.mdin
+  high_level_mdin: /path/to/high_level.mdin
+
+low_level: MNDOD
+high_level: PBE0
+# cutoff: QM-MM cutoff radius
+cutoff: 6.
+
+# do not change the value below
+sys_format: amber/rst7
+init_multi_systems: true
+model_devi_clean_traj: false
+model_devi_engine: amber
+model_devi_skip: 0
+shuffle_poscar: false
+fp_style: amber/diff
+detailed_report_make_fp: true
+use_clusters: true

setup.py

@@ -17,7 +17,7 @@
 with open(path.join('dpgen', '_date.py'), 'w') as fp :
     fp.write('date = \'%s\'' % today)
 
-install_requires=['numpy>=1.14.3', 'dpdata>=0.2.6', 'pymatgen>=2019.1.13', 'ase', 'monty>2.0.0', 'paramiko', 'custodian','GromacsWrapper>=0.8.0', 'dpdispatcher>=0.3.11']
+install_requires=['numpy>=1.14.3', 'dpdata>=0.2.6', 'pymatgen>=2019.1.13', 'ase', 'monty>2.0.0', 'paramiko', 'custodian','GromacsWrapper>=0.8.0', 'dpdispatcher>=0.3.11', 'netCDF4']
 
 setuptools.setup(
     name=NAME,

tests/generator/amber/ml.mdin

@@ -0,0 +1,77 @@
+title
+&cntrl
+! IO =======================================
+      irest = 1       ! 0 = start, 1 = restart
+        ntx = 5       ! 1 = start, 5 = restart
+       ntxo = 1       ! read/write rst as formatted file
+      iwrap = 1       ! wrap crds to unit cell
+     ioutfm = 1       ! write mdcrd as netcdf
+       imin = 0
+      ntmin = 1
+       ntpr = @freq@
+       ntwr = @freq@
+       ntwx = @freq@
+       ntwf = 0
+! DYNAMICS =================================
+     nstlim = @nstlim@   ! number of time steps
+         dt = 0.001   ! ps/step
+        ntb = 1       ! 1=NVT periodic, 2=NPT periodic, 0=no box
+! TEMPERATURE ==============================
+      temp0 = 298     ! target temp
+   gamma_ln = 5.0     ! Langevin collision freq
+        ntt = 3       ! thermostat (3=Langevin)
+! PRESSURE  ================================
+        ntp = 0       ! 0=no scaling, 1=isotropic, 2=anisotropic
+! SHAKE ====================================
+        ntc = 2       ! 1=no shake, 2=HX constrained, 3=all constrained
+noshakemask = "@qm_region@"    ! do not shake these
+        ntf = 1       ! 1=cpt all bond E, 2=ignore HX bond E, 3=ignore all bond E
+! MISC =====================================
+        cut = 9.0
+      ifqnt = 1
+         ig = -1
+     nmropt = 1
+/
+
+&wt
+type='DUMPFREQ', istep1=25
+&end
+&wt
+  type='END',
+&end
+DISANG=TEMPLATE.disang
+DUMPAVE=TEMPLATE.dumpave
+
+&ewald
+  dsum_tol = 1.e-6
+/
+
+&qmmm
+    qm_theory   = '@qm_theory@'
+        qmmask  = '@qm_region@'
+      qmcharge  = @qm_charge@
+          spin  = 1
+       qmshake  = 0
+      qm_ewald  = 1
+   qmmm_switch  = 1
+       scfconv  = 1.e-10
+     verbosity  = 0
+  tight_p_conv  = 1
+  diag_routine  = 0
+   pseudo_diag  = 1
+  dftb_maxiter  = 100
+/
+
+&ml
+  iml=1
+  mlmask="@qm_region@"
+  rcut = @rcut@
+  intrafile=""
+  alskip = @freq@
+  althresh = 1.e-6
+  interfile(1)="@GRAPH_FILE0@"
+  interfile(2)="@GRAPH_FILE1@"
+  interfile(3)="@GRAPH_FILE2@"
+  interfile(4)="@GRAPH_FILE3@"
+/
+

tests/generator/context.py

@@ -26,6 +26,7 @@
 param_pwmat_file = 'param-pyridine-pwmat.json'
 param_abacus_file = 'param-pyridine-abacus.json'
 param_abacus_post_file = 'param-methane-abacus.json'
+param_amber_file = "param-amber.json"
 
 def my_file_cmp(test, f0, f1):
     with open(f0) as fp0 :

tests/generator/param-amber.json

@@ -0,0 +1,54 @@
+{
+    "model_devi_jobs": [
+        {
+            "sys_idx": [0]
+        }
+    ],
+    "use_multi_systems": true,
+    "model_devi_engine": "amber",
+    "fp_style": "amber/diff",
+    "numb_models": 4,
+    "mdin_prefix": "amber",
+    "parm7_prefix": "amber",
+    "sys_prefix": "amber",
+    "disang_prefix": "amber",
+    "sys_configs": [
+        [
+            "init_-1.20.rst7"
+        ]
+    ],
+    "disang": [
+        "init_-1.20.disang"
+    ],
+    "parm7": [
+        "MON.parm7"
+    ],
+    "mdin": [
+        "ml.mdin"
+    ],
+    "qm_region": [
+        ":1"
+    ],
+    "qm_charge": [
+        0
+    ],
+    "r": [
+        [-1.20]
+    ],
+    "nsteps": [
+        1000
+    ],
+    "type_map": ["C", "H", "O", "N", "P", "HW", "OW"],
+    "shuffle_poscar": false,
+    "low_level": "DFTB2",
+    "high_level": "PBE0",
+    "cutoff": 6.0,
+    "fp_task_max": 1000,
+    "model_devi_skip": 0,
+    "model_devi_f_trust_lo": 0.10,
+    "model_devi_f_trust_hi": 0.25,
+    "fp_params": {
+        "low_level_mdin": "ml.mdin",
+        "high_level_mdin": "ml.mdin"
+    }
+}