Installation of QUIP and quippy

These intructions provide more details on the compilation and installation of QUIP (Fortran library and main programs) and quippy (Python interface).

Precompiled Containers

If you have access to Docker or Singularity, you can try one of the precompiled images to get up and running quickly.

Compilation Instructions

First try the quickstart below, which should work with most Linux systems. For Mac systems, have a look at Installing on Mac OS X with macports first.

Quick start

Install [1] the prerequisites: GCC, gfortran, Python, and the linear algebra libraries. For example, on Ubuntu, do (in a terminal):

$ sudo apt-get install gcc gfortran python python-pip libblas-dev liblapack-dev

For other systems, replace the apt-get part with your system package manager. Beware that the packages might also have slightly different names; these can usually be found with a quick search.

Don’t forget the quippy prerequisites:

$ pip install numpy
$ pip install ase

Now you can get the code and compile:

$ git clone --recursive https://github.com/libAtoms/QUIP.git
$ export QUIP_ARCH=linux_x86_64_gfortran
$ export QUIPPY_INSTALL_OPTS=--user  # omit for a system-wide installation
$ make config

Answer all the questions with their defaults (by pressing enter) for now, just to get things working.

$ make
$ make install-quippy

And now open a Python terminal and see if it works:

$ python
>>> import quippy
>>>

If the import completes successfully (i.e. with no output) then the installation was successful. You may want to continue with Installing the Jupyter notebook to run the interactive tutorials.

[1]	If this isn’t your machine and you don’t have root access, these packages might already be installed by the system administrator. If not, ask them.

Step by step

If that didn’t work, try these step-by-step instructions instructions excerpted from the top-level README. The README file is the most up-to-date source of installation information.

1. To compile QUIP the minimum requirements are:

   • A working Fortran compiler. QUIP is tested with gfortran 4.4 and later, and ifort 11.1.
   • Linear algebra libraries BLAS and LAPACK. QUIP is tested with reference versions libblas-dev and liblapack-dev on Ubuntu 12.04, and mkl 11.1 with ifort.

2. Clone the QUIP repository from GitHub. The --recursive option brings in submodules automatically (If you don’t do this, then you will need to run git submodule update --init from the top-level QUIP directory after cloning)

   git clone --recursive https://github.com/libAtoms/QUIP.git
   

3. Decide your architecture by looking in the arch/ directory, and define an environmental variable QUIP_ARCH, e.g.:

   export QUIP_ARCH=linux_x86_64_gfortran
   

   for standard gfortran on Linux. Here is where you can adjust which compiler is being used, if you do not like the defaults. You may need to create your own arch/Makefile.${QUIP_ARCH} file based on an existing file for more exotic systems.

4. Customise QUIP, set the maths libraries and provide linking options:

   make config
   

   Makefile.config will create a build directory, build/${QUIP_ARCH}, and all the building happen there. First it will ask you some questions about where you keep libraries and other stuff, if you don’t use something it is asking for, just leave it blank. The answers will be stored in Makefile.inc in the build/${QUIP_ARCH} directory, and you can edit them later (e.g. to change compiler, optimisation or debug options).

   If you later make significant changes to the configuration such as enabling or disabling tight-binding support you should force a full rebuild by doing a make deepclean; make.

5. Compile all programs, modules and libraries:

   make
   

   From the top-level QUIP directory. All programs are built in build/${QUIP_ARCH}/. You can also find compiled object files and libraries (libquip.a) in that directory. Programs can be called directly from that directory.

   Other useful make targets include:

   • make install : copies all compiled programs it can find to QUIP_INSTALLDIR, if it’s defined and is a directory (full path required), and copies bundled structures to QUIP_STRUCTS_DIR if it is defined.
   • make libquip: Compile QUIP as a library and link to it. This will make all the various libraries and combine them into one: build/${QUIP_ARCH}/libquip.a, which is what you need to link with (as well as LAPACK).

6. A good starting point is to use the quip program, which can calculate the properties of an atomic configuration using a variety of models. For example:

   quip at_file=test.xyz init_args='IP LJ' \
       param_file=share/Parameters/ip.parms.LJ.xml E
   

   assuming that you have a file called test.xyz with the following data in it representing Cu atoms in a cubic fcc lattice:

   4
   Lattice="3.61 0 0 0 3.61 0 0 0 3.61" Properties=species:S:1:pos:R:3
   Cu     0.000 0.000 0.000
   Cu     0.000 1.805 1.805
   Cu     1.805 0.000 1.805
   Cu     1.805 1.805 0.000
   

   The Lennard-Jones parameters in the above example are defined in the ip.parms.LJ.xml file under share/Parameters (ensure the path to this file is correct). The format of the atomic configuration is given in [Extended XYZ](http://libatoms.github.io/QUIP/io.html#extendedxyz) format, in which the first line is the number of atoms, the second line is a series of key=value pairs, which must at least contain the Lattice key giving the periodic bounding box and the Properties key that describes the remaining lines. The value of Properties is a sequence of triplets separated by a colon (:), that give the name, type and number of columns, with the type given by I for integers, R for reals, S for strings.

   Most string arguments can be replaced by --help and QUIP programs will then print a list of allowable keywords with brief help messages as to their usage, so e.g. init_args=--help will give a list of potential model types (and some combinations). The parsing is recursive, so init_args="IP --help" will then proceed to list the types of interatomic potentials (IP) that are available.

7. To compile the Python wrappers (quippy), the minimum requirements are:

   • Python 2.7. (Python 3 support is in progress).
   • [NumPy](http://www.numpy.org)
   • [Atomic Simulation Environment](https://wiki.fysik.dtu.dk/ase/) (ase)
   • (optional) [SciPy](http://www.scipy.org)
   • (optional) [matscipy](https://github.com/libAtoms/matscipy).

8. If you are using a Python virtual environment (virtualenv) and would like to install quippy into it, ensure the environment is activated (source <env_dir>/bin/activate, where <env_dir> is the root of your virtual environment) _before_ building quippy (otherwise library versions may cause unexpected conflicts).

9. To compile the Python wrappers (quippy), run:

   make quippy
   

   Quippy can be used by adding the lib directory in quippy/build/${QUIP_ARCH} to your $PYTHONPATH, however it can be more convenient to install into a specific Python distribution:

   make install-quippy
   

   will either install into the current virtualenv or attempt to install systemwide (usually fails without sudo). To install only for the current user (into ~/.local), execute the command QUIPPY_INSTALL_OPTS=--user make install-quippy, or use QUIPPY_INSTALL_OPTS=--prefix=<directory> to install into a specific directory. QUIPPY_INSTALL_OPTS can also be set in the file build/${QUIP_ARCH}/Makefile.inc.

10. More details on the quippy installation process and troubleshooting for

common build problems are available in the [online documentation](http://libatoms.github.io/QUIP/).

11. To run the unit and regression tests, which depend on quippy:

    make test
    

12. To get back to a state near to a fresh clone, use

    make distclean

13. Some functionality is only available if you check out other modules within the QUIP/src/ directories, e.g. the ThirdParty (DFTB parameters, TTM3f water model), GAP (Gaussian Approximation Potential models) and GAP-filler (Gaussian Approximation Potential model training). These packages are not distributed with QUIP because they come with different licensing restrictions, but you can get them [here](http://www.libatoms.org/gap/gap_download.html)

    GAP is a machine learning method that uses Gaussian process regression, and needs large data files to run. You can find potentials that have been published as well as training data in our [data repository](http://www.libatoms.org/Home/DataRepository).

14. In order to run QUIP potentials via LAMMPS, make libquip to get QUIP into library form, and then follow the instructions in the [LAMMPS documentation](http://lammps.sandia.gov/doc/pair_quip.html). You need at least 11 Aug 2017 version or later.

### Mac OS

We do not recommend Apple-shipped compilers and python, and we do not test compatibility with them. Either use MacPorts or Homebrew to obtain GNU compilers, and also use the python from there or Anaconda. As of this edit, gcc-8.1 produces as internal compiler error, but gcc-4.6 through to gcc-7 is fine.

If that still doesn’t work or you’re using a nonstandard architecture, try looking at Custom settings and Common Problems. As a last resort you can consult the issue tracker on Github.

Installing the Jupyter notebook

Jupyter is an environment for interactive computing that makes using Python much easier and more intuitive. Especially useful is its notebook environment, which provides a handy way to experiment with code, see the results, and have a record of your progress. The interactive getting-started tutorial is a Jupyter notebook that you can run and modify yourself.

To get Jupyter up and running, the following should suffice [2]:

$ pip install jupyter
$ jupyter notebook

This will open a new window in your browser that you can use to navigate through your filesystem. To access the interactive tutorials, you can run the jupyter notebook command from your QUIP/doc/Examples directory (or any enclosing directory) then navigate to the notebooks and open Introduction.ipynb to get started.

[2]	This assumes you’ve already run sudo apt-get install python-pip; pip install numpy; pip install ase as in the Quick start.

Custom settings

MATHS_LINKOPTS

Library options needed to link to BLAS and LAPACK libraries. Any working BLAS/LAPACK installation is fine. If you are using Linux, ATLAS is a good option, and you should use something like the following:

-L/usr/local/atlas -llapack -lf77blas -lcblas -latlas

On Mac OS X, there are build in LAPACK libraries in the Accelerate framework, which you can use by entering

-framework Accelerate

FOX_LIBDIR, FOX_INCDIR and FOX_LIBS

Directories containing FoX libraries and header files, and required link options. Should be read automatically from QUIP Makefiles.

QUIPPY_FCOMPILER

Fortran compiler to use. The shell command:

$ f2py -c --help-fcompiler

will print a list of detected compilers on your system. Use gnu95 for gfortran, intel for ifort on 32-bit platforms and intelem for ifort on 64-bit platforms.

QUIPPY_DEFINES Preprocessor macros which should be defined

when compiling quippy. Note that since the Fortran source files are preprocessed before being scanned by f90doc, it’s important to put all the -D options needed here and not in QUIPPY_F90FLAGS.

QUIPPY_F90FLAGS and QUIPPY_F77FLAGS

Extra flags to pass to Fortran 90 and 77 compilers

QUIPPY_OPT

Optimisation settings for Fortran compiler

QUIPPY_DEBUG

Set this to 1 to include debugging information in the compiled extension code. This also disables optimisation.

QUIPPY_CPP

Fortran preprocessor to use. Default is system cpp.

QUIPPY_INSTALL_OPTS

Installation options, e.g. specify --user to install for the current user --prefix=${PREFIX} to install in a non-default location.

QUIPPY_NO_TOOLS

If set to 1, omit compilation of extra tools such as the elasticity module.

QUIPPY_NO_CRACK

If set to 1, omit compilation of crack utilities.

HAVE_NETCDF4

Should be set to 1 to enable NetCDF4 support. Should be read automatically from QUIP.

NETCDF4_LIBS, NETCDF4_FLAGS

Linker flags for compiling with NetCDF4 support, and flags for finding header files. Should be read automatically from QUIP.

Common Problems

Permission errors when installing

If you are installing as root, you may need to make sure the value of the QUIP_ARCH gets through to the install script, e.g.

sudo QUIP_ARCH=darwin_x86_64_gfortran make install-quippy

Installing on Mac OS X with macports

Macports requires various packages to be installed to compile everything, and may require extra linking arguments. See the README.macports for the latest details.

RuntimeError when importing

If, after installing quippy, you get the error shown below when you try to import it for the first time, then you are a victim of a bug in early versions of Python 2.6.

>>> import quippy
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
 File "/home/ab686/QUIP/Tools/quippy/quippy/__init__.py", line 31, in
<module>
   _quippy.system.verbosity_push(0)
RuntimeError: more argument specifiers than keyword list entries
(remaining format:'|:_quippy.system.verbosity_push')

The solution is either to compile your own Python from the current svn snapshot, or to update numpy to workaround the fix. This can be done either by compiling numpy from source from an up-to-date svn snapshot, or by applying the patch manually.

ImportError when importing

If you get an ImportError with a message about unresolved dependancies then something went wrong with the linking process - check that all the libraries you’re linking against are correct. You can used ldd on Linux of otool -L on Mac OS X to check which libraries the _quippy.so Python extension is linked against.

Possible problems installing atomeye module

If you get an ImportError with a message ::

>>> import atomeye
ImportError: dlopen(/Users/silvia/lib/python/_atomeye.so, 2): Symbol not found: _Config_load_libatoms
Referenced from: /Users/silvia/lib/python/_atomeye.so
Expected in: flat namespace
in /Users/silvia/lib/python/_atomeye.so

be sure that you have set QUIP_ROOT variable before starting the compilation. If not make clean and recompile again

If you get an ImportError with a message ::

>>> import atomeye
ImportError: dlopen(/Users/silvia/lib/python/_atomeye.so, 2): Symbol not found: __gfortran_adjustl
Referenced from: /Users/silvia/lib/python/_atomeye.so
Expected in: flat namespace
in /Users/silvia/lib/python/_atomeye.so

be sure that the gfortran libraries are properly set in ATOMEYE_LIBS in Makefile.atomeye

Error compiling IPModel_GAP

If you get the following error during compilation:

/src/Potentials/IPModel_GAP.f95:51.22:

use descriptors_module
                      1
Fatal Error: Can't open module file 'descriptors_module.mod' for reading at (1): No such file or directory

The GAP_predict module is not publicly available, so the Makefile.inc must contain HAVE_GP_PREDICT = 0, and HAVE_GP_TEACH = 0.

Warning about quippy.castep when importing quippy

If you get the following warning message when importing quippy:

$ python
>>> from quippy import *
WARNING:root:quippy.castep import quippy.failed.

then don’t worry, the quippy.castep module is not redistributed with the main code. The rest of quippy works fine without it.

Internal compiler error with ifort

If you see an error like the following when using the Intel fortran compiler:

fortcom: Severe: **Internal compiler error: internal abort** Please
report this error along with the circumstances in which it occurred
in a Software Problem Report.
 Note: File and line given may not be explicit cause of this error.

ifort: error #10014: problem during multi-file optimization compilation (code 3)
backend signals

Then the problem is due to bugs in the compiler. As a workaround, setting QUIPPY_NO_CRACK =1 in Makefile.inc should solve the problem, at the cost of excluding the fracture utilities from quippy.

Linking error on Mac OS X

When recompiling quippy on top of a previous compilation, you may see errors like this:

collect2: ld returned 1 exit status ld: in
/QUIP/build.darwin_x86_64_gfortran/libquiputils.a, malformed
archive TOC entry for  ___elasticity_module_MOD_einstein_frequencies,
offset 1769103734 is beyond end of file 1254096

This seems to be a Mac OS X Lion problem with rebuilding static libraries (.a files). Removing the static libraries with rm ../../build.${QUIP_ARCH}/*.a and recompiling should solve the problem.

Segmentation Faults with OpenBLAS

The threading in OpenBLAS can interfere with the OpenMP resulting in segfaults. Either recompile OpenBLAS with USE_OPENMP=1 or disable threading with export OPENBLAS_NUM_THREADS=1 at runtime.