Merge pull request #179 from deepmodeling/develop

Merge: v2.3.4 updates

Author: dyzheng

CMakeLists.txt

@@ -20,6 +20,7 @@ option(USE_OPENMP " Enable OpenMP in abacus." ON)
 option(ENABLE_ASAN "Enable AddressSanitizer" OFF)
 option(BUILD_TESTING "Build ABACUS unit tests" OFF)
 option(GENERATE_TEST_REPORTS "Enable test report generation" OFF)
+option(INFO "Enable gathering of math library information" OFF)
 
 set(ABACUS_BIN_NAME abacus)
 list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/modules)
@@ -217,7 +218,7 @@ if(ENABLE_DEEPKS)
     HINTS ${libnpy_INCLUDE_DIR}
   )
   if(NOT libnpy_SOURCE_DIR)
-  include(FetchContent)
+    include(FetchContent)
     FetchContent_Declare(
       libnpy
       GIT_REPOSITORY https://github.com/llohse/libnpy.git
@@ -237,14 +238,22 @@ if(DEFINED Libxc_DIR)
   set(ENABLE_LIBXC ON)
 endif()
 if(ENABLE_LIBXC)
-  find_package(Libxc)
+  find_package(Libxc HINTS ${Libxc_DIR}/share/cmake/Libxc)
   if(${Libxc_FOUND})
     message("Using Libxc.")
     add_compile_definitions(USE_LIBXC)
     target_link_libraries(${ABACUS_BIN_NAME} Libxc::xc)
     include_directories(${Libxc_INCLUDE_DIRS})
   else()
-    message(WARNING "Will not use Libxc.")
+    include(FetchContent)
+    FetchContent_Declare(
+      Libxc
+      GIT_REPOSITORY https://gitlab.com/libxc/libxc.git
+      GIT_TAG "5.2.3"
+      GIT_SHALLOW TRUE
+      GIT_PROGRESS TRUE
+    )
+    FetchContent_MakeAvailable(Libxc)
   endif()
 endif()
 
@@ -261,6 +270,12 @@ add_compile_definitions(
   TEST_EXX_RADIAL=1
 )
 
+if(INFO)
+  message(STATUS "Will gather math lib info.")
+  add_compile_definitions(GATHER_INFO)
+  # modifications on blas_connector and lapack_connector
+endif()
+
 IF (BUILD_TESTING)
   set(CMAKE_CXX_STANDARD 14) # Required in orbital
   include(CTest)
@@ -310,7 +325,7 @@ target_link_libraries(${ABACUS_BIN_NAME}
     neighbor
     orb
     io
-    ions
+    relax
     lcao
     gint
     parallel
@@ -319,7 +334,7 @@ target_link_libraries(${ABACUS_BIN_NAME}
     pw
     ri
     driver
-    xc
+    xc_
     hsolver
     elecstate
     hamilt

docs/examples/dpgen.md

@@ -2,7 +2,9 @@
 
 [back to main page](../../README.md)
 
-[DP-GEN](https://github.com/deepmodeling/dpgen), the deep potential generator, is a package designed to generate deep learning based model of interatomic potential energy and force fields (Yuzhi Zhang, Haidi Wang, Weijie Chen, Jinzhe Zeng, Linfeng Zhang, Han Wang, and Weinan E, DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models, Computer Physics Communications, 2020, 107206). ABACUS can now interface with DP-GEN to generate deep potentials. In the following, we take the FCC aluminum as an example.
+[DP-GEN](https://github.com/deepmodeling/dpgen), the deep potential generator, is a package designed to generate deep learning based model of interatomic potential energy and force fields (Yuzhi Zhang, Haidi Wang, Weijie Chen, Jinzhe Zeng, Linfeng Zhang, Han Wang, and Weinan E, DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models, Computer Physics Communications, 2020, 107206). ABACUS can now interface with DP-GEN to generate deep potentials and performe autotests. In the following, we take the FCC aluminum as an example.
+
+## init_bulk and run
 
 This example can be found in examples/dpgen-example/init_and_run directory.
 
@@ -11,7 +13,7 @@ Firstly, one needs to prepare input files for ABACUS calculation, e.g., “INPUT
 Secondly, for the "dpgen init_bulk" step, an `init.json` file should be provided:
 
 
-```
+```json
 {
     "init_fp_style":    "ABACUS",   # abacus interface
     "stages":           [1,2,3,4],
@@ -36,7 +38,7 @@ Secondly, for the "dpgen init_bulk" step, an `init.json` file should be provided
 ```
 
 Next, for the "dpgen run" step, the following `run_param.json` should be provided.
-```
+```json
 {
     "type_map": [
         "Al"
@@ -181,6 +183,188 @@ Next, for the "dpgen run" step, the following `run_param.json` should be provide
 }
 ```
 
+## autotest
+
+This example can be found in examples/dpgen-example/autotest directory.
+
+`dpgen autotest` supports to perform `relaxation`,`eos` (equation of state),`elastic`,`surface`,`vacancy`, and `interstitial` calculations with ABACUS. A `property.json` and `machine.json` file need to be provided. For example,
+
+`property.json`:
+```json
+
+{
+    "structures":    ["confs/"],
+    "interaction": {
+        "type":         "abacus",
+        "incar":        "./INPUT",
+        "potcar_prefix":"./",
+        "potcars":      {"Al": "Al.PD04.PBE.UPF"},
+        "orb_files": {"Al":"Al_gga_10au_100Ry_3s3p2d.orb"}
+    },
+    "_relaxation": {
+            "cal_type": "relaxation",
+            "cal_setting":{
+                    "input_prop": "./INPUT.rlx"
+            }
+     },
+    "properties": [
+        {
+         "type":         "eos",
+         "vol_start":    0.85,
+         "vol_end":      1.15,
+         "vol_step":     0.01,
+         "cal_setting": {
+                         "relax_pos": true,
+                         "relax_shape": true,
+                         "relax_vol": false,
+                         "overwrite_interaction":{
+                                     "type": "abacus",
+                                     "incar": "./INPUT",
+                                     "potcar_prefix":"./",
+                                     "orb_files": {"Al":"Al_gga_10au_100Ry_3s3p2d.orb"},
+                                     "potcars": {"Al": "Al.PD04.PBE.UPF"} }
+                        }
+        },
+         {
+         "type":         "elastic",
+         "skip":         false,
+         "norm_deform":   1e-2,
+         "shear_deform":  1e-2
+        },
+        {
+         "type":         "vacancy",
+         "skip":         false,
+         "supercell":    [2, 2, 2]
+        },
+        {
+         "type":           "surface",
+         "skip":         true,
+         "min_slab_size":  15,
+         "min_vacuum_size":11,
+         "pert_xz":        0.01,
+         "max_miller":     3,
+         "cal_type":       "static"
+        }
+        ]
+}
+```
+
+`machine.json`
+
+```json
+{
+  "api_version": "1.0",
+  "deepmd_version": "2.1.0",
+  "train" :[
+    {
+      "command": "dp",
+      "machine": {
+        "batch_type": "DpCloudServer",
+        "context_type": "DpCloudServerContext",
+        "local_root" : "./",
+        "remote_profile":{
+          "email": "[email protected]",
+          "password": "xxx",
+          "program_id": 000,
+            "input_data":{
+                "api_version":2,
+                "job_type": "indicate",
+                "log_file": "00*/train.log",
+                "grouped":true,
+                "job_name": "Al-train-VASP",
+                "disk_size": 100,
+                "scass_type":"c8_m32_1 * NVIDIA V100",
+                "platform": "ali",
+                "image_name":"LBG_DeePMD-kit_2.1.0_v1",
+                "on_demand":0
+            }
+        }
+      },
+      "resources": {
+        "number_node":123473334635,
+        "local_root":"./",
+        "cpu_per_node": 4,
+        "gpu_per_node": 1,
+        "queue_name": "GPU",
+        "group_size": 1
+      }
+    }],
+  "model_devi":
+    [{
+      "command": "lmp -i input.lammps -v restart 0",
+      "machine": {
+        "batch_type": "DpCloudServer",
+        "context_type": "DpCloudServerContext",
+        "local_root" : "./",
+        "remote_profile":{
+          "email": "[email protected]",
+          "password": "xxx",
+          "program_id": 000,
+            "input_data":{
+              "api_version":2,
+              "job_type": "indicate",
+              "log_file": "*/model_devi.log",
+              "grouped":true,
+              "job_name": "Al-devia-ABACUS",
+              "disk_size": 200,
+              "scass_type":"c8_m32_1 * NVIDIA V100",
+              "platform": "ali",
+              "image_name":"LBG_DeePMD-kit_2.1.0_v1",
+              "on_demand":0
+            }
+        }
+      },
+      "resources": {
+        "number_node": 28348383,
+        "local_root":"./",
+        "cpu_per_node": 4,
+        "gpu_per_node": 1,
+        "queue_name": "GPU",
+        "group_size": 100
+      }
+    }],
+  "fp":
+    [{
+      "command": "OMP_NUM_THREADS=1 mpirun -np 16 abacus",
+      "machine": {
+        "batch_type": "DpCloudServer",
+        "context_type": "DpCloudServerContext",
+        "local_root" : "./",
+        "remote_profile":{
+          "email": "[email protected]",
+          "password": "xxx",
+         "program_id": 000,
+            "input_data":{
+              "api_version":2,
+              "job_type": "indicate",
+              "log_file": "task*/fp.log",
+              "grouped":true,
+              "job_name": "al-DFT-test",
+              "disk_size": 100,
+              "scass_type":"c32_m128_cpu",
+              "platform": "ali",
+              "image_name":"XXXXX",
+              "on_demand":0
+            }
+        }
+      },
+      "resources": {
+        "number_node": 712254638889,
+        "cpu_per_node": 32,
+        "gpu_per_node": 0,
+        "queue_name": "CPU",
+        "group_size": 2,
+        "local_root":"./",
+        "source_list": ["/opt/intel/oneapi/setvars.sh"]
+      }
+    }
+  ]
+}
+
+```
+
+For each property, the command `dpgen autotest make property.json` will generate the input files, `dpgen autotest run property.json machine.json` will run the corresponding tasks, and `dpgen autotest post property.json` will collect the final results. 
+
 
 Notes:
 - The ABACUS-DPGEN interface can be used in both pw and lcao basis.

docs/input-main.md

@@ -68,7 +68,7 @@
 
 - [Variables useful for debugging](#variables-useful-for-debugging)
 
-    [nurse](#nurse) | [t_in_h](#t_in_h) | [vl_in_h](#vl_in_h) | [vnl_in_h](#vnl_in_h) | [test_force](#test_force) | [test_stress](#test_stress) | [colour](#colour) | [test_just_neighbor](#test_just_neighbor)
+    [nurse](#nurse) | [t_in_h](#t_in_h) | [vl_in_h](#vl_in_h) | [vnl_in_h](#vnl_in_h) | [test_force](#test_force) | [test_stress](#test_stress) | [colour](#colour)
 
 - [DeePKS](#deepks)
 
@@ -148,6 +148,9 @@ This part of variables are used to control general system parameters.
   - *md*: molecular dynamics
   - *sto-scf*: do self-consistent electronic structure calculation with [stochastic DFT](#electronic-structure-sdft)
   - *sto-md*: molecular dynamics with [stochastic DFT](#electronic-structure-sdft)
+  - *test_memory* : checks memory required for the calculation. The number is not quite reliable, please use with care
+  - *test_neighbour* : only performs neighbouring atom search
+  - *gen_jle* : generates projectors for DeePKS; see also deepks_lmax_descriptor
 
   > Note: *istate* and *ienvelope* only work for LCAO basis set and are not working right now.
 - **Default**: scf
@@ -597,9 +600,9 @@ This part of variables are used to control the parameters of stochastic DFT (SDF
 - **Description**:
   - Different method to do SDFT.
   - 1: SDFT calculates $T_n(\hat{h})\ket{\chi}$ twice, where $T_n(x)$ is the n-th order Chebyshev polynomial and $\hat{h}=\frac{\hat{H}-\bar{E}}{\Delta E}$ owning eigen-value $\in(-1,1)$. This method cost less memory but slow.
-  - 2: SDFT calculates $T_n(\hat{h})\ket{\chi}$ once but need much more memory. This method is fast but when memory is not enough, only method 1 can be used.
-  - other: use 1
-- **Default**: 1
+  - 2: SDFT calculates $T_n(\hat{h})\ket{\chi}$ once but need much more memory. This method is much faster. Besides, it calculate $N_e$ with $\bra{\chi}\sqrt{\hat f}\sqrt{\hat f}\ket{\chi}$, which needs smaller [nche_sto](#nche_sto). However, when memory is not enough, only method 1 can be used.
+  - other: use 2
+- **Default**: 2
 
 #### nbands_sto
 
@@ -642,6 +645,12 @@ This part of variables are used to control the parameters of stochastic DFT (SDF
 - **Description**: Frequency (once each initsto_freq steps) to generate new stochastic orbitals when running md.
 - **Default**:1000
 
+#### npart_sto
+
+- **Type**: Integer
+- **Description**: Make memory cost to 1/npart_sto times of previous one when running post process of SDFT like DOS with method_sto = 2.
+- **Default**:1
+
 ### Geometry relaxation
 
 This part of variables are used to control the geometry relaxation.
@@ -963,8 +972,8 @@ Warning: this function is not robust enough for the current version. Please try
 #### deepks_descriptor_lmax
 
 - **Type**: Integer
-- **Description**: control the max angular momentum of descriptor basis.
-- **Default**: 0
+- **Description**: when generating projectors, this variable controls the max angular momentum of descriptor basis.
+- **Default**: 2
 
 #### deepks_scf
 
@@ -1597,12 +1606,6 @@ This part of variables are used to control berry phase and wannier90 interfacae
 - **Description**: If set to 1, output to terminal will have some color.
 - **Default**: 0
 
-#### test_just_neighbor
-
-- **Type**: Boolean
-- **Description**: If set to 1, then only perform the neighboring atoms search.
-- **Default**: 0
-
 ### Electronic conductivities
 
 Frequency-dependent electronic conductivities can be calculated with Kubo-Greenwood formula[Phys. Rev. B 83, 235120 (2011)].