Falicov-Kimball MC

Dependencies

• c++11-compatible compiler
• ALPSCore (http://alpscore.org)
• ALPSCore requirements : cmake, hdf5, mpi.
• Boost: boost-mpi
• arpack (arpack-ng)
• eigen http://eigen.tuxfamily.org/
• gtest : https://github.com/google/googletest (fetched automatically)

Installation

1. Get all dependencies.

2. Download fk_mc: git clone https://github.com/aeantipov/fk_mc.git.

   1. Create a temporary build directory
   2. In it, run

     cmake \
     -DCMAKE_INSTALL_PREFIX=FK_MC_INSTALL_DIR" \
     -DCMAKE_CXX_FLAGS="-ftemplate-depth=256" \
     -DTesting=ON \
     -DBenchmark=ON \
     -DCMAKE_BUILD_TYPE="Release" \
     -DALPSCore_DIR=path_to_ALPSCoreConfig.cmake_dir
     FK_MC_DOWNLOAD_DIR
   

   where FK_MC_DOWNLOAD_DIR is the location if the downloaded fk_mc source (with CMakeLists.txt), TRIQS_INSTALL_DIR is a location where you want to install fk_mc. 3. (optional for configuration) specify -DLATTICES="cubic2d;cubic3d;triangular" etc to compile the code for different lattice types and dimensions. By default all lattices are compiled. 4. run make 5. (optional) run make test and verify that the tests pass. This typically saves for most encountered errors. 6. run make install

Running fk_mc

In FK_MC_INSTALL_DIR/bin you will find fk_mc_LATTICE executables, where LATTICE corresponds to the chosen lattice type such as cubic2d. The standard way to run is to do

  mpirun fk_mc_cubic2d --beta 1.0 --U 1.0 --L 8

Here beta is the inverse temperature, U is the interaction strength, and L is the linear system size (volume = L^d). Run fk_mc_cubic2d --help to learn about all the parameters.

The output is stored in the hdf5 archive (by default - output.h5). Specifying --plaintext option in the run dumps the value of observables to plaintext files. For example specific heat is obtained then in cv_error.dat and reads: 5.000000e+01 -2.474323e-01 1.207943e-28 1.554312e-15. First column is the number of samples used, second - the value of the observable, third is variance, and last is the standard error. The binning is done to avoid autocorrelations, so the number of samples is typically smaller than the number of measurements.

Authors & Contributors

• Andrey Antipov, Andrey.E.Antipov[at]gmail.com, 2013-now.
• Andreas Herrmann, University of Fribourg.

Academic usage

• please cite http://arxiv.org/abs/1605.01390 [Phys. Rev. Lett. 117, 146601 (2016)].