DensPart is an atoms-in-molecules density partitioning program. At the moment, it only features one method to partition the density, namely the Minimal Basis Iterative Stockholder (MBIS) scheme. See http://dx.doi.org/10.1021/acs.jctc.6b00456
Disclaimer: This implementation is a prototype and is not extensively tested yet. Future revisions may break backward compatibility of the API and file formats.
Required dependencies:
- NumPy: https://numpy.org
- QC-Grid: https://github.com/theochem/grid
Install (with dependencies):
pip install git+https://github.com/theochem/grid.git
pip install git+https://github.com/theochem/denspart.git(There are no releases yet.)
One needs to construct a density.npz file, which is used as input for the denspart
script. (The optional dependencies below provide convenient tools to make such files.)
The file density.npz uses the NumPy ZIP (NPZ) format, which is a simple container file
format for arrays. More details on NPZ can be found here in the NumPy documentation. The file density.npz
should contain at least the following arrays:
points: Quadrature grid points, shape(npoint, 3).weights: Quadrature grid weights, shape(npoint, ).density: Electron density at the grid points, shape(npoint, ).atnums: Atomic numbers, shape(natom, ).atcoords: Nuclear coordinates, shape(natom, 3).cellvecs: (Optional) One, two or three cell vectors (rows) defining periodic boundary conditions, shape(nvec, 3).
All data are assumed to be in atomic units.
With a density.npz file, one can perform the partitioning as follows:
denspart density.npz results.npzThe output is stored in results.npz, and contains the following arrays. (These may
be subject to change in future code revisions.)
Copied from the input file
density.npz:atnums: Atomic numbers, shape(natom, ).atcoords: Nuclear coordinates, shape(natom, 3).
General outputs:
atnfn: The number of pro-density basis functions on each atom, shape(natom, ).atnpar: The number of pro-density parameters for each atom, shape(natom, ).charges: atomic partial charges, shape(natom,).multipole_moments: Multipole moments (using spherical harmonics), forlgoing from 1 to 4, shape(natom, (lmax + 1)**2 - 1). The moments are in HORTON 2 order.c10 c11 s11 c20 c21 s21 c22 s22 c30 c31 s31 c32 s32 c33 s33 c40 c41 s41 c42 s42 c43 s43 c44 s44
In this list, the prefix
cdenotes cosine-like real spherical harmonics andsdenotes the sine-like functions. The first digit refers to the degreeland the second to the orderm.propars: The pro-density parameters, shape(sum(atnpar), ).radial_moments: Expectation values ofr**n, forngoing from 0 to 4, shape(natom, 5).
MBIS-specific outputs:
core_charges: MBIS core charges, shape(natom,).valence_charges: MBIS valence charges, shape(natom,).valence_widths: MBIS valence widths, shape(natom,).
Algorithm settings:
gtol: A stopping condition that was used for the optimization of the pro-density parameters. This is a threshold on the gradient of the extended KL divergence.maxiter: A stopping condition that was used for the optimization of the pro-density parameters. This is the maximum number of iterations allowed in SciPy's trust-constr minimizer.density_cutoff: A density cutoff parameter that was used to determine the cutoff radii for the local integration grids.
The arrays in the results.npz file can be accessed in Python as follows:
import numpy as np
results = np.load("results.npz")
print("charges", results["charges"])
# From here, one can convert data to other formats:
# - CSV
np.savetxt("charges.csv", results["charges"], delimiter=",")
# - JSON
import json
json.dump(results["charges"].tolist(), open("charges.json", "w"))
# One can also easily post-process the results with some scripting:
# - Molecular dipole moment predicted by the atomic charges.
print(np.dot(results["atcoords"].T, results["charges"]))
# - Contribution to the molecular dipole moment due to the atomic dipoles.
# (This includes a reordering the spherical harmonics.)
print(results["multipole_moments"][:, [1, 2, 0]].sum(axis=0))See https://github.com/theochem/iodata
Install as follows:
pip install git+https://github.com/theochem/iodata.gitWhen IOData is installed, the npz output of the partitioning can be converted into an extended XYZ file as follows:
denspart-write-extxyz results.npz results.xyzIn order to derive a density.npz from several wavefunction file formats
(wfn, wfx, molden, fchk, ...), one needs install a two dependencies:
Install as follows:
pip install git+https://github.com/theochem/iodata.git
pip install git+https://github.com/theochem/gbasis.gitOnce these are installed, one can compute densities on a grid from a wavefunction file. For example:
denspart-from-horton3 some-file.fchk density.npzA minimal working example showing how to partition a density from a Gaussian FCHK can be found in examples/horton3.
One may also derive a density.npz file from a
GPAW calculation.
When GPAW is installed, one can run:
denspart-from-gpaw some-file.gpw density.npzA minimal working example can be found in examples/gpaw. Note that you may have to add mpirun in front of the command. However, the conversion does not yet support parallel execution and thus only works for the case of a single process, even when using mpirun.
One may also derive a density.npz from an ADF AMSJob.
When AMS is installed, you can install
denspart in the AMS Python environment as follows:
# If needed:
source ${ADFHOME}/amsbashrc.sh
# Avoid setting ADF and AMS environment variables manually, because these may change
# with different versions of AMS.
amspython -m pip install git+https://github.com/theochem/grid.git
amspython -m pip install git+https://github.com/theochem/denspart.git
# For writing the extended XYZ file:
amspython -m pip install git+https://github.com/theochem/iodata.gitThen, the conversion and partitioning are done as follows:
amspython -m denspart.adapters.adf ams.results density.npz
amspython -m denspart density.npz results.npz
amspython -m denspart.utils.write_extxyz results.npz results.xyzwhere ams.results is the directory with output files. You need to disable symmetry
and write out the TAPE10 file. More details can be found the the denspart.adapters.adf
module. A minimal working example can be found in examples/adf.
By adding a few lines to the Psi4 input script, it will write an NPZ file with Psi4's built-in molecular quadrature grids:
energy, wfn = psi4.energy(return_wfn=True)
from denspart.adapters.psi4 import write_density_npz
write_density_npz(wfn)Symmetry is not supported, so you need to set the point group to c1 when specifying
the geometry. A minimal working example can be found in examples/psi4.
The development environment is configured as follows. It is assumed that you have direnv installed. (If not, you can manually the virtual environment.)
# Clone git repo, assuming you have ssh access to github
git clone git@github.com:theochem/denspart.git
cd denspart
# Create a virtual environment with all dependencies needed for testing
python -m venv venv
cat > .envrc << 'EOL'
source venv/bin/activate
export GPAW_SETUP_PATH=${PWD}/venv/share/gpaw-setups-0.9.20000
EOL
direnv allow
pip install -U pip
pip install -e .
# Mandatory dependency, but not yet included in setup.py
pip install --upgrade git+https://github.com/theochem/grid.git
# Development tools
pip install --upgrade pre-commit ruff black
# Extra dependency for testing adapters
pip install --upgrade git+https://github.com/theochem/iodata.git
pip install --upgrade git+https://github.com/theochem/gbasis.git
pip install --upgrade git+https://github.com/tovrstra/pytest-regressions@npz
pip install --upgrade ase
# (Make sure BLAS and LibXC are installed, so GPAW can link to them.)
# Fedora: sudo dnf install libxc-devel blas-devel
pip install --upgrade gpaw
# Install GPAW pseudopotentials
gpaw install-data venv/shareTo run all tests locally:
pre-commit run --all
pytest