How to use mVMC?

Prerequisite

mVMC requires the following packages:

  • C compiler (intel, Fujitsu, GNU, etc. )

  • MPI library

  • Option: ScaLAPACK library (intel MKL, Fujitsu, ATLAS, etc.)

Note

Settings of intel compiler

When you use the intel compiler, you can use easily scripts attached to the compiler. In the case of the bash in 64 bit OS, write the following in your ~/.bashrc:

$ source /opt/intel/bin/compilervars.sh intel64

or

$ source /opt/intel/bin/iccvars.sh intel64
$ source /opt/intel/mkl/bin/mklvars.sh

Please read manuals of your compiler/library for more information.

Installation

You can download mVMC in the following place. https://github.com/issp-center-dev/mVMC/releases

You can obtain the mVMC directory by typing

$ tar xzvf mVMC-xxx.tar.gz

There are two kind of procedures to install mVMC.

Using mVMCconfig.sh

Please run mVMCconfig.sh script in the mVMC directory as follow (for ISSP system-B “sekirei”):

$ bash mVMCconfig.sh sekirei

Then environmental configuration file make.sys is generated in src/ directory. The command-line argument of mVMCconfig.sh is as follows:

  • sekirei : ISSP system-B “sekirei”

  • kei : K computer and ISSP system-C “maki”

  • intel-openmpi : Intel compiler + OpenMPI

  • intel-mpich : Intel compiler + MPICH2

  • intel-intelmpi : Intel compiler + IntelMPI

  • gcc-openmpi : GCC + OpenMPI

  • gcc-mpich-mkl : GCC + MPICH + MKL

make.sys is as follows (for ISSP-system-B “sekirei”):

CC = mpicc
F90 = mpif90
CFLAGS = -O3 -no-prec-div -xHost -qopenmp -Wno-unknown-pragmas
FFLAGS = -O3 -implicitnone -xHost
LIBS = -L $(MKLROOT)/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 \
        -lmkl_intel_thread -lmkl_core -lmkl_blacs_sgimpt_lp64 -lpthread -lm
SFMTFLAGS = -no-ansi-alias -DHAVE_SSE2

We explain macros of this file as:

  • CC : C compiler (mpicc, mpifccpx)

  • F90 : fortran compiler (ifort, frtpx)

  • Libs : Linker option

  • CFLAGS : C compile option

  • FFLAGS : fortran compile option

Then you are ready to compile mVMC. Please type

$ make mvmc

and obtain vmc.out and vmcdry.out in src/ directory; you should add this directory to the $PATH.

You can make a PATH to mVMC as follows: $ export PATH=${PATH}:mVMC_top_directory/src/ If you keep this PATH, you should write above in ~/.bashrc (for bash as a login shell)

Using cmake

We can compile mVMC as

cd $HOME/build/mvmc
cmake -DCONFIG=gcc $PathTomvmc
make

Here, we set a path to mVMC as $PathTomvmc and to a build directory as `` $HOME/build/mvmc``. After compilation, src directory is constructed below a $HOME/build/mvmc directory and we obtain an executable vmc.out in src/ directory.

In the above example, we compile mVMC by using a gcc compiler. We can select a compiler by using the following options:

  • sekirei : ISSP system-B “sekirei”

  • fujitsu : Fujitsu compiler

  • intel : Intel compiler + Linux PC

  • gcc : GCC compiler + Linux PC.

An example of compiling mVMC by using the Intel compiler is shown as follows:

mkdir ./build
cd ./build
cmake -DCONFIG=intel ../
make

After compilation, src/ directory is created below the build/ directory and an execute vmc.out in the src/ directory. We can select ScaLAPACK instead of LAPACK for vmc calculation by adding the following as the cmake option

-DUSE_SCALAPACK=ON -DSCALAPACK_LIBRARIES="xxx".

Here, xxx is the libraries to use ScaLAPACK. Please note that we must delete the build/ directory and repeat the above operations when we change the compiler.

Note

Before using cmake for sekirei, you must type

$ source /home/issp/materiapps/tool/env.sh

When we type the following in sekirei,

$ cmake -DCONFIG=sekirei ../ -DUSE_SCALAPACK=ON ,

-DSCALAPACK_LIBRARIES is automatically set as

-DSCALAPACK_LIBRARIES="\${MKLROOT}/lib/intel64 -lmkl_scalapack_lp64
-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core
-lmkl_blacs_sgimpt_lp64".

When the path to libraries for ScaLAPACK is different in your circumstance, please set -DSCALAPACK_LIBRARIES as the correct path.

Directory structure

When mVMC-xxx.tar.gz is unzipped, the following directory structure is composed.:

.
|-- CMakeLists.txt
|-- COPYING
|-- README.md
|-- config/
|   |-- fujitsu.cmake
|   |-- gcc.cmake
|   |-- intel.cmake
|   `-- sekirei.cmake
|-- dist.sh*
|-- doc/
|   |-- CMakeLists.txt
|   |-- bib/
|   |-- en/
|   |-- figs/
|   |-- ja/
|   |-- package/
|   `-- userguide.html
|-- mVMCconfig.sh
|-- samples/
|   |-- Standard/
|   |   |-- Hubbard/
|   |   |-- Kondo/
|   |   `-- Spin/
|   `-- Wannier/
|       |-- Sr2CuO3/
|       `-- Sr2VO4/
|-- src/
|   |-- ComplexUHF/
|   |-- StdFace/
|   |-- common/
|   |-- mVMC/
|   |-- pfapack/
|   `-- sfmt/
|-- test/
|   |-- CMakeLists.txt
|   |-- python/
|   `-- tool/
`-- tool/

Basic usage

mVMC works as whether the following two modes:

  • Expert mode

    mVMC supports the arbitrary fermion-/spin-lattice system; we can specify the hopping, etc. at each site independently. Although this makes us able to specify flexibly the target this requires many input-files, and the setup of the calculation is complicated.

  • Standard mode

    For some typical models (such as the Heisenberg model on the square lattice), we can start calculation with a few parameters (for example, the size of the simulation cell, the common coupling parameter). In this case, the input-files for Expert mode are automatically generated. Although the number of available systems is smaller than that number of Expert mode, the setup of the calculation is easier than in Expert mode.

We can calculate by using these modes as follows:

  1. Prepare a minimal input file

    You can choose a model (the Heisenberg model, the Hubbard model, etc.) and a lattice (the square lattice, the triangular lattice, etc.) from ones provided; you can specify some parameters (such as the first/second nearest neighbor hopping integrals, the on-site Coulomb integral, etc.) for them. The input file format is described in Input files for Standard mode.

  2. Run

    Run an executable vmc.out in terminal by specifying the name of input file written in previous step (option -s is required).

    $ mpiexec -np number-of-processes Path/vmc.out -s Input-file-name

    When you use a queuing system in workstations or super computers, sometimes the number of processes is specified as an argument for the job-submitting command. If you need more information, please refer manuals for your system.

  3. Watch calculation logs

    Log files are outputted in the output/ directory which is automatically made in the directory for a calculation scenario. The details of output files are shown in Output files.

  4. Results

    If the calculation is finished normally, the result files are outputted in the output/ directory. The details of output files are shown in Output files.

  5. Prepare and run Expert mode

    In the above case, the calculation starts as soon as input files for Expert mode are generated. If we only generate files without starting the calculation, we can use an executable vmcdry.out as follows (MPI is not used in this step):

    $ Path/vmcdry.out Input-file-name

    Then, we can edit generated files by hand and run a executable vmc.out with namelist.def as an argument (option -e is required) as follows:

    $ mpiexec -np number-of-processes Path/vmc.out -e namelist.def

Note

The number of threads for OpenMP

If you specify the number of OpenMP threads for mVMC, you should set it as follows (in case of 16 threads) before the running:

$ export OMP_NUM_THREADS=16

Printing version ID

By using -v option as follows, you can check which version of mVMC you are using.

$ PATH/vmcdry.out -v
$ PATH/vmc.out -v