The detailed installation processes for different environments are described below. The easiest installation with a good computation performance is achieved by using the phono3py conda package (see Installation instruction of latest development version of phono3py).

Installation of phonopy before the installation of phono3py is required. See how to install phonopy at Phono3py relies on phonopy, so please use the latest release of phonopy when installing phono3py.

Installation using conda

Using conda is the easiest way for installation of phono3py if you are using x86-64 linux system or macOS. These packages are made and maintained by Jan Janssen. The installation is simply done by:

% conda install -c conda-forge phono3py

All dependent packages should be installed.

Installation from source code

When installing phono3py using from the source code, a few libraries are required before running script.

For phono3py, OpenMP library is necessary for the multithreding support. In additon, BLAS, LAPACK, and LAPACKE are also needed. These packages are probably installed using the package manager for each OS or conda.

When using gcc to compile phono3py, libgomp1 is necessary to enable OpenMP multithreading support. This library is probably installed already in your system. If you don’t have it and you use Ubuntu linux, it is installed by:

% sudo apt-get install libgomp1

Installation of LAPACKE

LAPACK library is used in a few parts of the code to diagonalize matrices. LAPACK*E* is the C-wrapper of LAPACK and LAPACK relies on BLAS. Both single-thread or multithread BLAS can be used in phono3py. In the following, multiple different ways of installation of LAPACKE are explained.

MKL LAPACKE (with multithread BLAS)

Phono3py can be compiled with MKL for using LAPACKE. If finds the file named, the contents of is read and those are included in the compilation setting. For example, the following setting prepared as seems working on Ubuntu 16.04 system:

intel_root = "/opt/intel/composer_xe_2015.7.235"
mkl_root = "%s/mkl" % intel_root
compiler_root = "%s/compiler" % intel_root

mkl_extra_link_args_lapacke = ['-L%s/lib/intel64' % mkl_root,
mkl_extra_link_args_lapacke += ['-L%s/lib/intel64' % compiler_root,
mkl_include_dirs_lapacke = ["%s/include" % mkl_root]

This setting considers to use icc but it may be compiled with gcc. With gcc, the compiler related setting shown above (i.e., around compiler_root) is unnecessary. To achieve this installation, not only the MKL library but also the header file are necessary. The libraries are linked dynamically, so in most of the cases, LD_LIBRARY_PATH environment variable has to be correctly specified to let phono3py find those libraries.

OpenBLAS provided by conda (with multithread BLAS)

The installtion of LAPACKE is easy by conda. It is:

% conda install -c conda-forge openblas libgfortran

The recent change of openblas package provided from anaconda makes to install nomkl, i.e., numpy and scipy with Intel MKL cannot be used together with openblas. At this moment, this is avoided to install openblas from conda-forge channel. If the python libraries are not yet installed:

% conda install -c conda-forge numpy scipy h5py pyyaml matplotlib

Note that using hdf5 files on NFS mouted file system, you may have to disable file locking by setting


This openblas package contains BLAS, LAPACK, and LAPACKE. When this libopenblas is linked and the else statement of the C macro definition section in is executed, the following macro are activated:

if use_setuptools:
    extra_compile_args += ['-DMULTITHREADED_BLAS']
    define_macros += [('MULTITHREADED_BLAS', None)]

Libraries or headers are not found at the build by, the following setting may be of the help:

extra_link_args_lapacke += ['-lopenblas', '-lgfortran']
include_dirs_lapacke += [
    os.path.join(os.environ['CONDA_PREFIX'], 'include'), ]

Netlib LAPACKE provided by Ubuntu package manager (with single-thread BLAS)

In the versions of Ubuntu-12.10 or later, LAPACKE ( can be installed from the package manager (liblapacke and liblapacke-dev):

% sudo apt-get install liblapack-dev liblapacke-dev

Compiling Netlib LAPACKE

The compilation procedure is found at the LAPACKE web site. After creating the LAPACKE library, liblapacke.a (or the dynamic link library), must be properly modified to link it. As an example, the procedure of compiling LAPACKE is shown below.

% tar xvfz lapack-3.6.0.tgz
% cd lapack-3.6.0
% cp
% make lapackelib

BLAS, LAPACK, and LAPACKE, these all may have to be compiled with -fPIC option to use it with python.

Building using

If package installation is not possible or you want to compile with special compiler or special options, phono3py is built using In this case, manual modification of may be needed.

  1. Download the latest source code at

  2. and extract it:

    % tar xvfz phono3py-
    % cd phono3py-

    The other option is using git to clone the phonopy repository from github:

    % git clone
    % cd phono3py
  1. Set up C-libraries for python C-API and python codes. This can be done as follows:

    Run script via pip:

    % pip install -e .
  2. Set $PATH and $PYTHONPATH

    PATH and PYTHONPATH are set in the same way as phonopy, see

Installation instruction of latest development version of phono3py

When using conda, PYTHONPATH should not be set if possible because potentially wrong python libraries can be imported.

This installation instruction supposes linux x86-64 environment.

  1. Download miniconda package

    Miniconda is downloaded at

    For usual 64-bit Linux system:

    % wget -O ~/

    For macOS:

    % wget -O ~/

    The installation is made by

    % bash ~/ -b -p $HOME/miniconda
    % export PATH="$HOME/miniconda/bin:$PATH"

    The detailed installation instruction is found at To use the conda-forge channel as the default one, ~/.condarc may be written following the conda-forge official documentation (

  2. Initialization of conda and setup of conda environment

    % conda init <your_shell>

    <your_shell> is often bash but may be something else. It is important that after running conda init, your shell is needed to be closed and restarted. See more information by conda init --help.

    Then conda allows to make conda installation isolated by using conda’s virtual environment.

    % conda create -n phono3py -c conda-forge python=3.8
    % conda activate phono3py

    Use of this is strongly recommended, otherwise proper settings of CONDA_PREFIX, C_INCLUDE_PATH, and LD_LIBRARY_PATH will be necessary.

  1. Install necessary conda packages for phono3py

    % conda install -c conda-forge numpy scipy h5py pyyaml matplotlib-base compilers "libblas=*=*mkl" spglib mkl-include

    Note that using hdf5 files on NFS mouted file system, you may have to disable file locking by setting


    Install the latest phonopy and phono3py from github sources:

    % mkdir dev
    % cd dev
    % git clone
    % git clone
    % cd phonopy
    % git checkout develop
    % python build
    % pip install -e .
    % cd ../phono3py
    % git checkout develop
    % python build
    % pip install -e .

    The conda packages dependency can often change and this recipe may not work properly. So if you find this instruction doesn’t work, it is very appreciated if letting us know it in the phonopy mailing list.

Multithreading and its controlling by C macro

Phono3py uses multithreading concurrency in two ways. One is that written in the code with OpenMP parallel for. The other is achieved by using multithreaded BLAS. The BLAS multithreading is depending on which BLAS library is chosen by users and the number of threads to be used may be controlled by the library’s environment variables (e.g., OPENBLAS_NUM_THREADS or MKL_NUM_THREADS). In the phono3py C code, these two are written in a nested way, but of course the nested use of multiple multithreadings has to be avoided. The outer loop of the nesting is done by the OpenMP parallel for code. The inner loop calls LAPACKE functions and then the LAPACKE functions call the BLAS routines. If both of the inner and outer multithreadings can be activated, the inner multithreading must be deactivated at the compilation time. This is achieved by setting the C macro MULTITHREADED_BLAS, which can be written in Deactivating the multithreading of BLAS using the environment variables is not recommended because it is used in the non-nested parts of the code and these multithreadings are unnecessary to be deactivated.

Trouble shooting

  1. Phonopy version should be the latest to use the latest phono3py.

  2. There are other pitfalls, see