Thanks to some recent, excellent work by Sebastian Ehlert, NTPoly is now available from Conda. Conda is a cross platform package and environment manager. One of its strengths is that you can setup many virtual environments, ensuring that there are no issues with version clashes between packages. I use it extensively myself for python development, I really like that I can setup a fresh environment whenever I need.

To use NTPoly from conda, create/activate an environment:

conda create --name ntpoly-from-conda
conda activate ntpoly-from-conda

Of course you can use a preexisting environment as well. Then you can install NTPoly:

conda install -c conda-forge ntpoly

The NTPoly package does not come with any compilers, so we will need to install those too.

conda install -c conda-forge compilers

To test it out, I recommend downloading the files in the PremadeMatrix example that comes with NTPoly. We now will create a Makefile that is able to build those files using the NTPoly installed by conda.

ALL: fortran cxx
CC=mpicxx
FC=mpif90
CXXFLAGS=-O3 -fopenmp
FFLAGS=-O3 -fopenmp
FLIBS=-lNTPoly
CXXLIBS=-lNTPolyCPP -lNTPolyWrapper -lNTPoly
TESTARGS= --process_rows 1 --process_columns 1 --process_slices 1 \
          --hamiltonian Hamiltonian.mtx --overlap Overlap.mtx \
          --number_of_electrons 10 --threshold 1e-6 \
          --converge_overlap 1e-3 --converge_density 1e-5 \
          --density Density.mtx

fortran: main.f90
        $(FC) $(FFLAGS) $< -o $@ $(FLIBS)
cxx: main.cc
        $(CC) $(CXXFLAGS) $< -o $@ $(CXXLIBS)

test:
        mpirun -np 1 ./fortran $(TESTARGS)
        mpirun -np 1 ./cxx $(TESTARGS)

clean:
        rm fortran cxx

The code should now compile with no issue using the make command. CMake also works out of the box, you can download the CMakeLinkage example to start. Then the usual mkdir build ; cd build ; cmake .. ; make should work with no issues.

There is one outstanding issue with this approach that appears if you try to use the C++ bindings. This is related to a bug on MPICH; you can see the discussion on the MPICH Feedstock. Hopefully this is resolved soon, but in the meantime the conda distribution of NTPoly is a great way to get started using the Fortran bindings.