Molecular response properties in implicit and explicit solvent environments
---------------------------------------------------------------------------

These datasets contain structures, energies, forces and various response properties
(dipole moment, polarizability tensor, nuclear shielding tensors) for ethanol, 
methanol and allyl-p-tolyl ether used for training the FieldSchNet model. Two
ethanol datasets are computed in implicit solvent environments using polarizable
continuum models. Datasets of molecular properties computed in the presence of a
electric field generated by explicit point charges (for QM/MM) are available for
ethanol and allyl-p-tolyl ether. Atomic units are used for all quantities. All
calculations have been performed at the PBE0/def2-TZVP level of theory [1,2] using
the ORCA quantum chemistry package (version 4.0.1.2)[3]. For details on how
individual datasets were generated and specific computational settings please
refer to the publication

Format
------

The datasets are stored as tar archives compressed via gzip and can be unpacked
on a Unix machine using the the "tar -xzf <dataset>.tgz" command.

The data itself is stored in form of an atomic simulation environment (ASE)
sqlite database[4]. The database can be accessed using the standard tools
provided by ASE. In addition, a utility class (schnetpack.data.AtomsData) for
loading the datasets can be found in the SchNetPack code package for atomistic
machine learning [5]

The following example python code can be used to load the data associated with
a single entry with the index <idx> from a database file:

""" 
from schnetpack.data import AtomsData

database = AtomsData("path/to/db")
atoms, properties = database.get_properties(idx)
"""

<atoms> is an ASE Atoms object containing the positions and atom types, while
<properties> is a dictionary containing all stored properties. Properties 
include energy (“energy”), forces (“forces”), dipole moments (“dipole moment”),
polarizability tensors (“polarizability”) and, in most cases, nuclear shielding
tensors (“shielding”). All quantities use Hartree atomic units.

The following datasets are available:
 - ethanol_vacuum.tgz: ethanol in vacuum, all properties available.
 - ethanol_continuum_evw.tgz: ethanol in vacuum (ε=1.0), ethanol (ε=24.3) and
       water (ε=80.4) computed with a polarizable continuum model. The dielectric
       constant is stored in the entry “dielectric_constant“. All properties are
       available.
 - ethanol_continuum_mt.tgz: ethanol in methanol (ε=32.63) and toluene (ε=10.3)
       computed with a polarizable continuum model. The dielectric constant is
       stored in the entry “dielectric_constant“. All properties are available.
 - ethanol_qmmm.tgz: ethanol structures in fields of external point charges used
       for QM/MM models. The electric field computed for the environment is
       stored in the entry “electric_field“.  All properties are available. 
 - ethanol_methanol_qmmm.tgz: ethanol and methanol structures identified with
       adaptive sampling. All structures embedded in fields of points charges.
       The ethanol structures are a subset selected from the ethanol_qmmm.db
       dataset. Only energy, forces, dipole moment and polarizabilities are
       available. Free atom energies of the individual elements are provided in
       the metadata.
 - ate_vacuum.tgz: allyl-p-tolyl ether in vacuum, all properties available.
 - ate_qmmm.tgz: allyl-p-tolyl ether structures in fields of external point
       charges used for QM/MM models. The electric field computed for the
       environment is stored in the entry “electric_field“. All properties are
       available. 

How to cite
-----------

M. Gastegger, K. T. Schütt, K. R. Müller Machine learning of solvent effects
   on molecular spectra and reactions, Chem. Sci., 10.1039/D1SC02742E (2021).

References and links
---------------------
[1] Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 77 (18),
    3865–3868 (1996). 
[2] Weigend, F.; Ahlrichs, R. Phys. Chem. Chem. Phys. 7, 3297-3305 (2005).
[3] Neese, F. Wiley Interdiscip. Rev. Comput. Mol. Sci. 2, 73-78 (2012).
[4] https://wiki.fysik.dtu.dk/ase/index.html
[5] https://github.com/atomistic-machine-learning/schnetpack