Merge branch 'develop' into io_libxc

Author: mohanchen

.github/workflows/test.yml

@@ -39,7 +39,7 @@ jobs:
             --from-ref ${{ github.event.pull_request.base.sha }}
               --to-ref ${{ github.event.pull_request.head.sha }}
         continue-on-error: true
-      - uses: pre-commit-ci/lite-action@v1.1.0
+      - uses: pre-commit-ci/lite-action@v1.0.3
 
       - name: Build
         run: |

CMakeLists.txt

@@ -39,6 +39,7 @@ option(ENABLE_RAPIDJSON "Enable rapid-json usage." OFF)
 option(ENABLE_CNPY "Enable cnpy usage." OFF)
 option(ENABLE_PEXSI "Enable support for PEXSI." OFF)
 option(ENABLE_CUSOLVERMP "Enable cusolvermp." OFF)
+option(USE_DSP "Enable DSP usage." OFF)
 
 # enable json support
 if(ENABLE_RAPIDJSON)
@@ -119,6 +120,12 @@ elseif(ENABLE_LCAO AND NOT ENABLE_MPI)
   set(ABACUS_BIN_NAME abacus_serial)
 endif()
 
+if (USE_DSP)
+  set(USE_ELPA OFF)
+  set(ENABLE_LCAO OFF)
+  set(ABACUS_BIN_NAME abacus_dsp)
+endif()
+
 list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 
 if(ENABLE_COVERAGE)
@@ -240,6 +247,11 @@ if(ENABLE_MPI)
   list(APPEND math_libs MPI::MPI_CXX)
 endif()
 
+if (USE_DSP)
+  target_link_libraries(${ABACUS_BIN_NAME} ${DIR_MTBLAS_LIBRARY})
+  add_compile_definitions(__DSP)
+endif()
+
 find_package(Threads REQUIRED)
 target_link_libraries(${ABACUS_BIN_NAME} Threads::Threads)
 
@@ -436,22 +448,6 @@ if(ENABLE_FLOAT_FFTW)
 endif()
 
 if(ENABLE_DEEPKS)
-  # Torch uses outdated components to detect CUDA arch, causing failure on
-  # latest CUDA kits. Set CMake variable TORCH_CUDA_ARCH_LIST in the form of
-  # "major.minor" if required.
-  find_package(Torch REQUIRED)
-  if(NOT Torch_VERSION VERSION_LESS "2.1.0")
-    set_if_higher(CMAKE_CXX_STANDARD 17)
-  elseif(NOT Torch_VERSION VERSION_LESS "1.5.0")
-    set_if_higher(CMAKE_CXX_STANDARD 14)
-  endif()
-  include_directories(${TORCH_INCLUDE_DIRS})
-  if(MKL_FOUND)
-    list(PREPEND math_libs ${TORCH_LIBRARIES})
-  else()
-    list(APPEND math_libs ${TORCH_LIBRARIES})
-  endif()
-  add_compile_options(${TORCH_CXX_FLAGS})
   target_link_libraries(${ABACUS_BIN_NAME} deepks)
 
   find_path(libnpy_SOURCE_DIR npy.hpp HINTS ${libnpy_INCLUDE_DIR})
@@ -470,6 +466,25 @@ if(ENABLE_DEEPKS)
   add_compile_definitions(__DEEPKS)
 endif()
 
+# Torch uses outdated components to detect CUDA arch, causing failure on
+# latest CUDA kits. Set CMake variable TORCH_CUDA_ARCH_LIST in the form of
+# "major.minor" if required.
+if(ENABLE_DEEPKS OR DEFINED Torch_DIR)
+  find_package(Torch REQUIRED)
+  if(NOT Torch_VERSION VERSION_LESS "2.1.0")
+    set_if_higher(CMAKE_CXX_STANDARD 17)
+  elseif(NOT Torch_VERSION VERSION_LESS "1.5.0")
+    set_if_higher(CMAKE_CXX_STANDARD 14)
+  endif()
+  include_directories(${TORCH_INCLUDE_DIRS})
+  if(MKL_FOUND)
+    list(PREPEND math_libs ${TORCH_LIBRARIES})
+  else()
+    list(APPEND math_libs ${TORCH_LIBRARIES})
+  endif()
+  add_compile_options(${TORCH_CXX_FLAGS})
+endif()
+
 if (ENABLE_CNPY)
   find_path(cnpy_SOURCE_DIR
     cnpy.h
@@ -590,13 +605,14 @@ if(DEFINED DeePMD_DIR)
     add_compile_definitions(__DPMDC)
   else()
     target_link_libraries(${ABACUS_BIN_NAME} DeePMD::deepmd_cc)
-    if(NOT DEFINED TensorFlow_DIR)
-      set(TensorFlow_DIR ${DeePMD_DIR})
-    endif()
-    find_package(TensorFlow REQUIRED)
-    if(TensorFlow_FOUND)
-      target_link_libraries(${ABACUS_BIN_NAME} TensorFlow::tensorflow_cc)
-    endif()
+  endif()
+endif()
+
+if(DEFINED TensorFlow_DIR)
+  find_package(TensorFlow REQUIRED)
+  include_directories(${TensorFlow_DIR}/include)
+  if(TensorFlow_FOUND)
+    target_link_libraries(${ABACUS_BIN_NAME} TensorFlow::tensorflow_cc)
   endif()
 endif()
 

docs/advanced/elec_properties/Berry_phase.md

@@ -23,7 +23,7 @@ pseudo_dir      ../../../tests/PP_ORB  //the path to locate the pesudopotential
 orbital_dir     ../../../tests/PP_ORB  //the path to locate the numerical orbital files
 ntype         3
 ecutwfc       50 // Ry
-symmetry      0 // turn off symmetry
+symmetry      -1 // turn off symmetry
 calculation   nscf // non-self-consistent calculation
 basis_type    lcao // atomic basis
 init_chg  file // read charge from files
@@ -70,4 +70,6 @@ The results are shown as follows:
  P =    0.8906925  (mod    2.1748536)  (   0.0000000,   0.0000000,   0.8906925) C/m^2
 ```
 
-The electric polarization **P** is multivalued, which modulo a quantum e**R**/V~cell~. Note: the values in parentheses are the components of the **P** along the c axis in the x, y, z Cartesian coordinates when set gdir = 3 in INPUT file.
+The electric polarization **P** is multivalued, which modulo a quantum e**R**/V~cell~. 
+
+Note: The vectors R1, R2, and R3 refer to the three lattice vectors of the unit cell. When gdir=3, the calculated polarization is along the R3 direction. The three values in parentheses represent the re-projection of the polarization along the R3 direction onto the Cartesian coordinate system (i.e., the xyz coordinate system). To obtain the full polarization components in the Cartesian system, you need to calculate the polarization for R1, R2, and R3 separately, and then sum their respective x, y, and z components.

docs/advanced/input_files/input-main.md

@@ -39,7 +39,6 @@
     - [pw\_diag\_thr](#pw_diag_thr)
     - [pw\_diag\_nmax](#pw_diag_nmax)
     - [pw\_diag\_ndim](#pw_diag_ndim)
-    - [diago\_full\_acc](#diago_full_acc)
     - [erf\_ecut](#erf_ecut)
     - [fft\_mode](#fft_mode)
     - [erf\_height](#erf_height)
@@ -779,12 +778,6 @@ These variables are used to control the plane wave related parameters.
 - **Description**: Only useful when you use `ks_solver = dav` or `ks_solver = dav_subspace`. It indicates dimension of workspace(number of wavefunction packets, at least 2 needed) for the Davidson method. A larger value may yield a smaller number of iterations in the algorithm but uses more memory and more CPU time in subspace diagonalization.
 - **Default**: 4
 
-### diago_full_acc
-
-- **Type**: bool
-- **Description**: Only useful when you use `ks_solver = dav_subspace`. If `TRUE`, all the empty states are diagonalized at the same level of accuracy of the occupied ones. Otherwise the empty states are diagonalized using a larger threshold (10-5) (this should not affect total energy, forces, and other ground-state properties).
-- **Default**: false
-
 ### erf_ecut
 
 - **Type**: Real
@@ -925,7 +918,7 @@ calculations.
   - **cg**: cg method.
   - **bpcg**: bpcg method, which is a block-parallel Conjugate Gradient (CG) method, typically exhibits higher acceleration in a GPU environment.
   - **dav**: the Davidson algorithm.
-  - **dav_subspace**: subspace Davidson algorithm
+  - **dav_subspace**: Davidson algorithm without orthogonalization operation, this method is the most recommended for efficiency. `pw_diag_ndim` can be set to 2 for this method.
 
   For atomic orbitals basis,
 
@@ -3964,6 +3957,13 @@ Currently supported: `RPA`, `LDA`, `PBE`, `HSE`, `HF`.
 - **Description**:  The number of 2-particle states to be solved
 - **Default**: 0
 
+### lr_unrestricted
+- **Type**: Boolean
+- **Description**: Whether to use unrestricted construction for LR-TDDFT (the matrix size will be doubled).
+  - True:  Always use unrestricted LR-TDDFT. 
+  - False: Use unrestricted LR-TDDFT only when the system is open-shell.
+- **Default**: False
+
 ### abs_wavelen_range
 
 - **Type**: Real Real

docs/advanced/input_files/kpt.md

@@ -8,7 +8,7 @@ ABACUS uses periodic boundary conditions for both crystals and finite systems. F
 
 ## Gamma-only Calculations
 
-In ABACUS, we offer th option of running gamma-only calculations for LCAO basis by setting [gamma_only](./input-main.md#gamma_only) to be 1. Due to details of implementation, gamma-only calculation will be slightly faster than running a non gamma-only calculation and explicitly setting gamma point to be the only the k-point, but the results should be consistent.
+In ABACUS, we offer the option of running gamma-only calculations for LCAO basis by setting [gamma_only](./input-main.md#gamma_only) to be 1. Due to details of implementation, gamma-only calculation will be slightly faster than running a non gamma-only calculation and explicitly setting gamma point to be the only the k-point, but the results should be consistent.
 
 > If gamma_only is set to 1, the KPT file will be overwritten. So make sure to turn off gamma_only for multi-k calculations.
 
@@ -23,7 +23,7 @@ method to generate k-mesh, and the following is an example input k-point (`KPT`)
 K_POINTS //keyword for start
 0 //total number of k-point, `0' means generate automatically
 Gamma //which kind of Monkhorst-Pack method, `Gamma' or `MP'
-2 2 2 0 0 0 //first three number: subdivisions along recpri. vectors
+2 2 2 0 0 0 //first three number: subdivisions along reciprocal vectors
             //last three number: shift of the mesh
 ```
 
@@ -63,8 +63,8 @@ Direct //`Direct' or `Cartesian' coordinate
 ## Band structure calculations
 
 ABACUS uses specified high-symmetry directions of the Brillouin zone for band structure
-calculations. The third line of k-point file should start with ‘Line’ or ‘Line_Cartesian’ for
-line mode. ‘Line’ means the positions below are in Direct coordinates, while ‘Line_Cartesian’
+calculations. The third line of k-point file should start with 'Line' or 'Line_Cartesian' for
+line mode. 'Line' means the positions below are in Direct coordinates, while 'Line_Cartesian'
 means in Cartesian coordinates:
 
 ```

docs/advanced/install.md

@@ -38,13 +38,22 @@ cmake -B build -DENABLE_DEEPKS=1 -DTorch_DIR=~/libtorch/share/cmake/Torch/ -Dlib
 If the Deep Potential model is employed in Molecule Dynamics calculations, the following prerequisites and steps are needed:
 
 - [DeePMD-kit](https://github.com/deepmodeling/deepmd-kit)
-- [TensorFlow](https://www.tensorflow.org/)
+- [TensorFlow](https://www.tensorflow.org/) (optional)
+- [LibTorch](https://pytorch.org/) (optional)
 
+In the simplest case, the `tensorflow_cc` and `torch` libraries are in the same directory as the `deepmd_c`/`deepmd_cc` libraries, then
 ```bash
-cmake -B build -DDeePMD_DIR=~/deepmd-kit -DTensorFlow_DIR=~/tensorflow
+cmake -B build -DDeePMD_DIR=/dir_to_deepmd-kit
 ```
+DeePMD-kit supports TensorFlow backend but its libraries are placed at another directory, then
 
-> `deepmd_c`/`deepmd_cc` and `tensorflow_cc` libraries would be called according to `DeePMD_DIR` and `TensorFlow_DIR`, which is showed in detail in [this page](https://github.com/deepmodeling/deepmd-kit/blob/master/doc/inference/cxx.md). If `TensorFlow_DIR` is not defined, it will be the same as `DeePMD_DIR`. Note that `tensorflow_cc` is not required if `deepmd_c` is found.
+```bash
+cmake -B build -DDeePMD_DIR=/dir_to_deepmd-kit -DTensorFlow_DIR=/dir_to_tensorflow
+```
+Similarly, DeePMD-kit supports PyTorch backend but its libraries are placed at another directory, then
+```bash
+cmake -B build -DDeePMD_DIR=/dir_to_deepmd-kit -DTorch_DIR=/dir_to_pytorch
+```
 
 ## Build with LibRI and LibComm
 
@@ -282,15 +291,21 @@ directly.
 
 > Note: This part is only required if you want to load a trained DeeP Potential and run molecular dynamics with that. To train the DeeP Potential with DP-GEN, no extra prerequisite is needed and please refer to [this page](http://abacus.deepmodeling.com/en/latest/advanced/interface/dpgen.html) for ABACUS interface with DP-GEN.
 
-To compile ABACUS with DeePMD-kit, you need to define `DeePMD_DIR` and `TensorFlow_DIR` in the file `Makefile.vars` or use
+To compile ABACUS with DeePMD-kit, you need to define `DeePMD_DIR` and `TensorFlow_DIR` (TensorFlow Backend, optional) and/or `LIBTORCH_DIR` (PyTorch Backend, optional) in the file `Makefile.vars`.
 
+Or the `tensorflow_cc` and `torch` libraries are in the same directory as the `deepmd_c`/`deepmd_cc` libraries, then
 ```makefile
-make DeePMD_DIR=~/deepmd-kit TensorFlow_DIR=~/tensorflow
+make DeePMD_DIR=/dir_to_deepmd-kit
 ```
+DeePMD-kit supports TensorFlow backend but its libraries are placed at another directory, then
 
-directly.
-
-> `deepmd_c`/`deepmd_cc` and `tensorflow_cc` libraries would be called according to `DeePMD_DIR` and `TensorFlow_DIR`, which is showed in detail in [this page](https://github.com/deepmodeling/deepmd-kit/blob/master/doc/inference/cxx.md). If `TensorFlow_DIR` is not defined, it will be the same as `DeePMD_DIR`. Note that `tensorflow_cc` is not required if `deepmd_c` is found.
+```makefile
+make DeePMD_DIR=/dir_to_deepmd-kit TensorFlow_DIR=/dir_to_tensorflow
+```
+Similarly, DeePMD-kit supports PyTorch backend but its libraries are placed at another directory, then
+```makefile
+make DeePMD_DIR=/dir_to_deepmd-kit Torch_DIR=/dir_to_pytorch
+```
 
 ### Add LibRI Support
 To use new EXX, you need two libraries: [LibRI](https://github.com/abacusmodeling/LibRI) and [LibComm](https://github.com/abacusmodeling/LibComm) and need to define `LIBRI_DIR` and `LIBCOMM_DIR` in the file `Makefile.vars` or use

docs/advanced/pp_orb.md

@@ -19,7 +19,20 @@ Inside ABACUS, orbitals in LCAO basis are arranged lexicographically by species-
 
 ## Generating atomic orbital bases
 
-Users may also choose to generate their own atomic obitals. In ABACUS, the atomic orbital bases are generated using a scheme developed in the [paper](https://iopscience.iop.org/article/10.1088/0953-8984/22/44/445501). A detailed description of the procedure for generating orbitals will be provided later.
+Users may also generate ABACUS numerical atomic obitals based on their own flavor. The theoretical background of orbital generation can be found in following works:
+
+- Spillage: [Chen M, Guo G C, He L. Systematically improvable optimized atomic basis sets for ab initio calculations[J]. Journal of Physics: Condensed Matter, 2010, 22(44): 445501.](https://iopscience.iop.org/article/10.1088/0953-8984/22/44/445501)
+- PTG DPSI: [Lin P, Ren X, He L. Strategy for constructing compact numerical atomic orbital basis sets by incorporating the gradients of reference wavefunctions[J]. Physical Review B, 2021, 103(23): 235131.](https://journals.aps.org/prb/abstract/10.1103/PhysRevB.103.235131)
+
+Guidelines for generating atomic orbital bases are as follows:
+
+- [Numerical Atomic Orbitals 1: the nomenclature and usage of numerical atomic orbitals in ABACUS](https://mcresearch.github.io/abacus-user-guide/abacus-nac1.html) (Chinese)
+- [Numerical Atomic Orbitals 2: generate numerical atomic orbitals based on given norm-conserving pseudopotential](https://mcresearch.github.io/abacus-user-guide/abacus-nac1.html) (Chinese)
+- [Numerical Atomic Orbitals 3: generate high-precision numerical atomic orbitals](https://mcresearch.github.io/abacus-user-guide/abacus-nac1.html) (Chinese)
+
+Stable orbital generation programs can be found in guidelines above, there is also another developing version of orbital generation program, in which algorithms are consecutively improved: [Github repository of ABACUS ORBGEN project](https://github.com/kirk0830/ABACUS-ORBGEN), the usage of which can be found in README (in English) file.
+
+*NOTE*: users are encouraged to cite the above works when numerical atomic orbitals and its generation codes are used in their research.
 
 ## BSSE Correction
 
@@ -51,16 +64,46 @@ $$
 $$
 
 ## Pseudopotentials
-
-In ABACUS, we support norm-conserving and ultrasoft pseudopotentials. 
-For norm-conserving pseudopotentials, we support four different formats of the pseudopotential files: UPF, UPF2, VWR, and BLPS. 
-For ultrasoft pseudopotentials, currently we support only one format of the pseudopotential files: UPF2.
-
-For more information, check the `ATOMIC_SPECIES` section in the specification of the [STRU file](./input_files/stru.md).
-
-Here we list some common sources of the pseudopotential files:
-
-1. [Quantum ESPRESSO](http://www.quantum-espresso.org/pseudopotentials/).
-2. [SG15-ONCV](http://quantum-simulation.org/potentials/sg15_oncv/upf/).
-3. [DOJO](http://www.pseudo-dojo.org/).
-4. [BLPS](https://github.com/PrincetonUniversity/BLPSLibrary).
+### Supported formats
+ABACUS supports both norm-conserving and ultrasoft pseudopotentials. For norm-conserving pseudopotentials, UPF, UPF2, VWR, and BLPS formats are supported. For ultrasoft pseudopotentials, UPF and UPF2 formats are supported. 
+
+### Usage
+For more information about pseudopotential usage, check the `ATOMIC_SPECIES` section in the specification of the [STRU file](./input_files/stru.md).
+
+### Download
+Users can find pseudopotentials in the following links:
+
+**Website**
+- [Quantum ESPRESSO](https://www.quantum-espresso.org/pseudopotentials): the official website of Quantum ESPRESSO, where you can find a large number of pseudopotential files.
+- [Stantard Solid State Pseudopotential library](https://www.materialscloud.org/sssp): a library of **high-quality** pseudopotentials for solid-state calculations, with **a large number of tests on efficiency and precison**.
+- [PWmat](http://www.pwmat.com/potential-download): a website that provides a large number of pseudopotential files, various kinds of semi-core constructed pseudopotentials are included. **Several sets (with or without f-electrons/noncolinear core correction) of Lanthanide pseudopotentials are also available**.
+- [THEOS](http://theossrv1.epfl.ch/Main/Pseudopotentials): PSlibrary 0.3.1, a library of pseudopotentials for DFT calculations, including ultrasoft, paw, norm-conserving both full-relativistic and scalar-relativistic pseudopotentials.
+- [ABACUS@USTC](https://abacus.ustc.edu.cn/pseudo/list.htm): **ABACUS official website** where you can find a large number of pseudopotential files and numerical atomic orbital files.
+- [BLPS](https://github.com/PrincetonUniversity/BLPSLibrary): BLPS format pseudopotential library
+
+**Norm-conserving pseudopotentials**
+- [SG15](http://www.quantum-simulation.org/potentials/sg15_oncv/): **vastly used in ABACUS** DFT calculation and numerical atomic orbital generation.
+- [PseudoDOJO](http://www.pseudo-dojo.org/): another widely used pseudopotential database, developed by Abinit group, **including Lanthanide pseudopotentials (f-electrons frozen)**.
+- [The Rappe group](https://www.sas.upenn.edu/rappegroup/research/pseudo-potential-gga.html): a collection of GGA pseudopotentials which are generated with Opium code, several tests proves that are out-performing in alloy systems.
+- [Matteo Giantomassi's Github repo](https://github.com/gmatteo/pseudos_ac_she): a Github repository that contains norm-conserving pseudopotentials for **Actinides and superheavy elements to 120-th element**.
+
+**Ultrasoft pseudopotentials**
+- [Vanderbilt](http://www.physics.rutgers.edu/~dhv/uspp/): a collection of ultrasoft pseudopotentials generated by Vanderbilt group.
+- [GBRV](https://www.physics.rutgers.edu/gbrv/) by Kevin F. Garrity, Joseph W. Bennett, Karin M. Rabe, and David Vanderbilt: presently the most popular ultrasoft pseudpotentials in Quantum ESPRESSO user community.
+
+### Pseudopotential Generation
+For pseudopotential generation, please refer to the following links for more information:
+- [Quantum ESPRESSO](http://www.quantum-espresso.org/pseudopotentials/)
+- [ONCVPSP](http://www.mat-simresearch.com/)
+- [Opium](https://opium.sourceforge.net/)
+
+A Chinese guideline is also available here: [A brief introduction of norm-conserving pseudopotential generation](https://mcresearch.github.io/abacus-user-guide/abacus-upf.html)
+
+# ABACUS Pseudopotential-Numerical atomic orbital Square (APNS) project
+For the purpose of providing high-quality pseudopotentials and numerical atomic orbitals, we have initiated the APNS project. The project is aimed at providing a large number of high-quality pseudopotentials and numerical atomic orbitals, along with diverse test data for the ABACUS user community, reduce the cost of generating and testing pseudopotentials and numerical atomic orbitals by users, and promote the development of ABACUS software. The project is currently in the development stage, and we welcome contributions from the community. For more information, please refer to the following links:
+- [APNS website: test data and results](https://kirk0830.github.io/ABACUS-Pseudopot-Nao-Square/)
+- [APNS workflow (Github repository): high-throughput test of pseudopotentials and numerical atomic orbitals](https://github.com/kirk0830/ABACUS-Pseudopot-Nao-Square)
+
+There are also other excellent projects that provide high-quality pseudopotentials along with test data:
+- [Solid State Pseudopotential library](https://www.materialscloud.org/sssp)
+- [Verification of the precision of DFT implementation via AiiDA common workflows](https://acwf-verification.materialscloud.org/)