0.24.0 / 2021-11-18

New Features

  • (GH#275) AtomicResult gains a native_files field of a dictionary of file names (or generic 'input') and text (not binary) contents that the CMS program may have generated but which haven’t necessarily been harvested into QCSchema. Contents controlled by a new native_files protocol analogous to stdout protocol.


  • (GH#281) TorsionDriveInput now accepts a list of Molecule objects as the initial_molecule to seed the torsiondrive with multiple conformations.

0.23.0 / 2021-09-23

Breaking Changes

  • (GH#276) AtomicResultProperties.dict() no longer forces arrays to JSON flat lists but now allows NumPy arrays. That is, AtomicResultProperties now behaves like every other QCElemental model. Expected to be disruptive to QCFractal.

  • (GH#280) Examples of QCSchema from test cases are now saved at branch . These have passed validation as models by Pydantic and as JSON by schema generated from Pydantic models.

New Features


  • (GH#274) The molecule from_string parser when no dtype specified learned to return the most specialized error message among the dtypes, not the full input string.

  • (GH#276) Molecule.to_string(..., dtype="nwchem") learned to handle ghosts (real=False) correctly. It also now prints the user label, which is used downstream for custom basis sets and shows up in a NWChem output file. QCEngine will be edited to process the label, but other uses may need modification.

  • (GH#276) Molecule.align learned a new keyword generic_ghosts=True so that it can act on molecules that have centers with content Gh, not Gh(He).

Bug Fixes

  • (GH#276) Molecule.to_string(..., dtype="gamess") learned to handle ghosts (real=False) correctly for coord=unique. Note that QCEngine uses coord=prinaxis, so actual ghosts are still NOT interpretable by downstream GAMESS.

0.22.0 / 2021-08-26

New Features

  • (GH#268) Add provisional models that store the inputs to and outputs of a torsion drive procedure. @SimonBoothroyd

  • (GH#272) Add SCF and return gradient and Hessian fields to AtomicResultProperties.


  • (GH#271) Molecule learned to create instances with geometry rounded to other than 8 decimal places through Molecule(..., geometry_noise=<13>) to optionally override qcel.models.molecule.GEOMETRY_NOISE = 8. This should be used sparingly, as it will make more molecules unique in the QCA database. But it is sometimes necessary for accurate finite difference steps and to preserve intrinsic symmetry upon geometry rotation. Previous route was to reset the qcel module variable for the duration of instance creation.

  • (GH#271) Molecule.align and Molecule.scramble learned to return a fuller copy of self than previously. Now has aligned atom_labels, real, and mass_numbers as well as incidentals like Identifiers. Fragmentation still not addressed.

  • (GH#271) Molecule.to_string(dtype="gamess") learned to write symmetry information to the prinaxis output if passed in through field fix_symmetry. This is provisional, as we’d like the field to be uniform across qcprogs.

Bug Fixes

  • (GH#271) Testing function compare_values() on arrays corrected the RMS maximum o-e value displayed and added a relative value.

0.21.0 / 2021-06-30

New Features

  • (GH#267) Serialization learned msgpack mode that, in contrast to msgpack-ext, doesn’t embed NumPy objects.


  • (GH#266) Testing function compare_values() learned to print RMS and MAX statistics for arrays.

Bug Fixes

0.20.0 / 2021-05-16

New Features

  • (GH#257) PhysicalConstantsContext learned to write a Fortran header. @awvwgk


  • (GH#261) Documentation became type-aware and grew more links.

0.19.0 / 2021-03-04

New Features


Bug Fixes

  • (GH#253) Fixed incompatibility with Pydantic >=1.8.

0.18.0 / 2021-02-16

New Features

  • (GH#237) Exports models to JSON Schema with make schema.

  • (GH#237) Build bank of JSON examples from Pydantic models defined in tests. Test that bank against exported schema with pytest --validate qcelemental/.

  • (GH#237) Many model descriptions edited, dimensions added to array properties, AtomicInput.model.basis now takes BasisSet object not just string, several properties added to match QCSchema, several limitations on number and uniqueness imposed.


  • (GH#237) Improve mypy conformance including dynamic provenance. Necessitates Pydantic >=1.5.

  • (GH#237) a0 without underscore added as computable pint unit.

  • (GH#246, GH#250) Removes types deprecated in NumPy v1.20.0.

Bug Fixes

  • (GH#244) Fixes where in code validation throws if center_data missing from BasisSet model.

  • (GH#249) Fixes web tests that weren’t marked.

0.17.0 / 2020-10-01


  • (GH#238) molparse.to_string MRChem dtype developed.

0.16.0 / 2020-08-19

New Features


  • (GH#231) compare, compare_values, and compare_recursive learned new keyword equal_phase that when True allows either computed or -computed to pass the comparison. For compare_recursive, the leniency can be restricted to specific leaves of the iterable by passing a list of allowed leaves.

Bug Fixes

  • (GH#229) Molecule.align told the full truth in its documentation that the mol_align argument can take a float.

0.15.1 / 2020-06-25

Bug Fixes

  • (GH#228) Fix testing bug for installed module, which was missing a dummy directory.

0.15.0 / 2020-06-25

New Features

  • (GH#182) Added experimental protocol for controlling autocorrection attemps. (That is, when a calculation throws a known error that QCEngine thinks it can tweak the input and rerun.) Currently in trial for NWChem.


  • (GH#186, GH#223) molparse.to_string Orca and MADNESS dtypes developed.

  • (GH#226) Allow which_import to distinguish between ordinary and namespace packages.

  • (GH#227) Add non-default strict argument to periodictable.to_Z, to_symbol, and to_element that fails when isotope information is given.

  • (GH#227) Allow nonphysical masses to pass validation in molparse.from_schema(..., nonphysical=True). Also allowed in forming qcel.models.Molecule(..., nonphysical=True).

Bug Fixes

  • (GH#227) Fixed deception described in issue 225 where qcel.models.Molecule(..., symbols=["O18"]) accepted “O18” but did not influence the isotope, as user might have expected. That now raises NotAnElementError, and an example of correctly setting isotope/masses has been added. This error now caught at qcel.molparse.from_arrays so general.

0.14.0 / 2020-03-06

New Features


  • (GH#211) Improve testing reliability by excusing PubChem when internet flaky.

  • (GH#216) “CODATA2018” constants now tested.

  • (GH#207) Multipoles exist in AtomicResultProperties as ndarray with order-dimensional shape. Property scf_quadrupole_moment defined.

Bug Fixes

  • (GH#216) Fixes a bug where “CODATA2018” constants could not be used with conversion_factor.

  • (GH#217) Can now run .schema() on pydantic classes containing Array fields (allowing ndarray in place of List).

0.13.1 / 2020-02-05

New Features

  • (GH#209) Added the option to Hill-order molecular formulas.

Bug Fixes

  • (GH#208) Fixes a Molecule hashing issue due to order of operations changes in Molecule.from_data. The order of operations changed in Molecule.from_data and occasionally resulted in different hashes for Molecules undergoing orient operations. This issue was introduced in 0.13.0 and is unlikely to have any serious negative effects as this did not affect hash integrity.

0.13.0 / 2020-01-29

New Features

  • (GH#183, GH#187) Added metadata about DFT functionals (

  • (GH#184) Optional PubChem identifiers were added to molecules.

  • (GH#187, GH#192, GH#195) Added metadata about CPUs (


  • (GH#179, GH#181) QCElemental works with Python 3.8 at the expense of needing a new 0.10 pint (rather than generic install). Pint 0.10 has optional NumPy dependency of >=1.12.0, so QCElemental that requires both NumPy and pint needs this constraint.

  • (GH#172, GH#173, GH#202, GH#203) Contribution improvements, including GitHub Actions checking, make format guidance, and updated

  • (GH#189, GH#196) Constants and unit conversion based on 2018 CODATA are available (but 2014 remains the default).

  • (GH#197, GH#199, GH#200) Added more atomic units and aliases (e.g. au_length = bohr).

  • (GH#190, GH#191, GH#201) Slim molecules. Many fields in Molecule objects may be optionally inferred.

Bug Fixes

  • (GH#174) Fix compare_recursive for when expected is a list but computed is not.

  • (GH#177) Spelling and type hint fixes.

  • (GH#180) Better test coverage.

0.12.0 / 2019-11-13

New Features


  • (GH#156) Molecules can now be correctly compared with ==.

  • (GH#157) molparse.to_string Q-Chem dtype developed. Psi4 dtype now includes label and doesn’t have extraneous info for single fragment systems.

  • (GH#162) New protocol stdout added to ResultProtocols controlling whether stdout field (which generally contains the primary logfile, whether a program writes it to file or stdout) is returned.

  • (GH#165) The code base is now Black formatted.


  • (GH#156) is deprecated and will be removed in v0.13.0.

  • (GH#167, GH#168) ResultInput, Result, Optimization have been removed in favor of AtomicInput, AtomicResult, and OptimizationResult and will be removed in v0.13.0.

Bug Fixes

  • (GH#170) ProtoModel subclasses now correctly allow custom __repr__ and __str__ methods.

  • (GH#164, GH#166) nglview-sdf molecule string format now correctly uses correct sdf format widths fixing some issues with very large molecules.

0.11.1 / 2019-10-28

Bug Fixes

  • (GH#152) Patches Molecule.from_file and Molecule.from_data to read XYZ+ format and correctly handle keyword arguments.

    Patches Molecule.to_file to write XYZ+ format as the default for XYZ and XYZ+ files.

0.11.0 / 2019-10-24


  • (GH#147) Updates Pydantic to the 1.0 release and fixes a number of breaking changes.

  • (GH#148) Switches from Py3dMoljs to NGLView for molecular visualization due to Jupyter Widget integration.

  • (GH#149) Adds statC and Debye to the units registry.

Bug Fixes

  • (GH#150) Patches which_import to correctly handle submodules.

0.10.0 / 2019-10-16


  • (GH#144) Allows which_import to handle submodules.

  • (GH#143) Allow testing complex numbers.

0.9.0 / 2019-10-01

New Features

  • (GH#137, GH#138) Coordinates can now be output in Turbomole format in addition to all other formats.

  • (GH#139) A wavefunction property have been added to the Result Model. Adds the ability for Engine and other programs to store and fetch wavefunction data.

  • (GH#140) Protocols have been added to QCInputSpecification which allows data to pre-pruned by different specifications. Main intention is to reduce wavefunction data which may be re-computed cheaply rather than storing all of it. This does change the input model, so requires a minor version bump.


  • (GH#132) BasisSet and Result’s documentations have been brought up to the standards of other models.

0.8.0 / 2019-09-13

New Features

  • (GH#123) QCElemental now passes MyPy!

  • (GH#127, GH#131) Adds van der Waals radii data available through vdwradii.get(atom) function.


Bug Fixes

  • (GH#125) Add back a consistency check that had been optimized out.

0.7.0 / 2019-08-23


  • (GH#118) Model string representations should be more user friendly and descriptive without overload the output.

  • (GH#119) The molparse.to_string keyword-arg return_data now returns molecule keywords for GAMESS and NWChem. The models.Molecule.to_string can use return_data now, too.

  • (GH#120) Auto documentation tech is now built into the ProtoModel and does not need an external function.

0.6.1 / 2019-08-19

Bug Fixes

  • (GH#114) The Numpy einsum calls reference the top level functions and not core C functions. This fixes an issue which can result in NumPy version dependencies.

0.6.0 / 2019-08-14

New Features

  • (GH#85, GH#87) Msgpack is a new serialization option for Models. Serialization defaults to msgpack when available (conda install msgpack-python [-c conda-forge]), falling back to JSON otherwise. This results in substantial speedups for both serialization and deserialization actions and should be a transparent replacement for users within Elemental itself.


  • (GH#78) Molecular alignments can now be aligned on the derivatives of vector components.

  • (GH#81) Testing is now operated both on the minimal supported and the latest released versions of dependencies.

  • (GH#82) Molecule fragment grouping is now disabled by default to match expected behavior.

  • (GH#84) Testing without internet connection should now pass since PubChem testing is skipped with no connection.

  • (GH#85) Molecule switches from lists to numpy arrays for internal storage of per-atom fields.

  • (GH#86) Molecule performance and memory enhancements through reduced validation times and LRU caching of common validations.

  • (GH#88, GH#109) The Molecule Model now has its attributes documented and in an on-the-fly manner derived from the Pydantic Schema of those attributes.

  • (GH#99, GH#100, GH#101, GH#102, GH#103, GH#104, GH#105, GH#106, GH#107) Various documentation, type hints, and small changes.

Bug Fixes

  • (GH#87) Molecule objects built from Schema are run through validators for consistency.

0.5.0 / 2019-07-16


  • (GH#76) Adds a built-in Molecule.to_file function for easy serialization into .numpy, .json, .xyz, .psimol, and .psi4 file formats.

Bug Fixes

  • (GH#74) Atom and fragment ordering are preserved when invoking get_fragment.

0.4.2 / 2019-06-13

New Features

  • (GH#70, GH#72) molparse.to_string Molpro dtype developed.

0.4.1 / 2019-05-31

New Features


  • (GH#68) molparse.to_string learned parameter return_data that contains aspects of the models.Molecule not expressible in the string. Implemented for dtypes xyz, cfour, psi4.

  • (GH#68) Datum gained an attribute numeric that reflects whether arithmetic on data is valid. Datums that aren’t numeric can now be created by initializing with numeric=False.

Bug Fixes

  • (GH#66) Fix tests when networkx not installed.

  • (GH#67) Fix “unsupported format string passed to numpy.ndarray.__format__” on Mac for testing.compare_values.

0.4.0 / 2019-05-13

New Features

  • (GH#51) Changes models.Molecule connectivity to default to None rather than an empty list. WARNING this change alters the hashes produced from the Molecule.get_hash functionality.

  • (GH#52, GH#53) models.Molecule learned nuclear_repulsion_energy, nelectrons, and to_string functions.

  • (GH#54) models.ResultProperties supports CCSD and CCSD(T) properties.

  • (GH#56) Algorithms Kabsch molutil.kabsch_align, Hungarian util.linear_sum_assignment, and Uno added. Utilities to generate random 3D rotations util.random_rotation_matrix and reindex a NumPy array into smaller blocks util.blockwise_expand added.

  • (GH#56) Molecular alignment taking into account displacement, rotation, atom exchange, and mirror symmetry for superimposable and differing geometries was added in molutil.B787 (basis NumPy function) and models.Molecule.align (far more convenient). Suitable for QM-sized molecules. Requires addition package networkx.

  • (GH#58) utils learned which_import and which that provide a path or boolean result for locating modules or commands, respectively. These were migrated from QCEngine along with safe_version and parse_version to colocate the import utilities.

  • (GH#61) Add molecular visualization to the models.Molecule object through the optional 3dMol.js framework.

  • (GH#65) testing.compare_molrecs learned parameter relative_geoms='align' that lets Molecules pass if geometries within a translation and rotation of each other.

  • (GH#65) testing.compare_recursive learned parameter forgive that is a list of paths that may differ without failing the comparison.


  • (GH#52, GH#53) molparse.to_string NWChem and GAMESS dtypes developed.

  • (GH#57) molparse.to_string learned dtype='terachem' for writing the separate XYZ file required by TeraChem. Angstroms or Bohr allowed, though the latter requires extra in input file.

  • (GH#60) util.which added the Python interpreter path to the default search $PATH.

  • (GH#62) Added * to parameter list of many functions requiring subsequent to be keyword only. Code relying heavily on positional arguments may get broken.

  • (GH#63) util.which learned parameter env to use an alternate search $PATH.

  • (GH#63) util.which and util.which_import learned parameters raise_error and raise_msg which raises ModuleNotFoundError (for both functions) when not located. It error will have a generic error message which can be extended by raise_msg. It is strongly encouraged to add specific remedies (like how to install) through this parameter. This is the third exit pattern possible from the “which” functions, of which path/None is the default, True/error happens when raise_error=True, and True/False happens otherwise when return_bool=True.

  • (GH#65) Testing functions compare, compare_values, compare_recursive learned parameter return_handler that lets other printing, logging, and pass/fail behavior to be interjected.

Bug Fixes

  • (GH#63) util.which uses os.pathsep rather than Linux-focused :.

  • (GH#65) Fixed some minor printing and tolerance errors in molecule alignment.

  • (GH#65) testing.compare_recursive stopped doing atol=10**-atol for atol>=1, bringing it in line with other compare functions.

0.3.3 / 2019-03-12


  • (GH#49) Precompute some mass number and mass lookups and store on qcel.periodic_table. Also move static re.compile expressions out of fns on to module. Mol validation .127s –> .005s.

0.3.2 / 2019-03-11

New Features

  • (GH#47) models.DriverEnum now has a derivative_int function to return 1 for gradient, etc., for easy math. properties returns 0.

  • (GH#47) Optional fix_symmetry field in qcschema_molecule was missing from models.Molecule so Pydantic got mad at Psi4. Now calmed.


  • (GH#48) If Molecule object has passed through molparse validation because it was created with a molparse constructor (e.g., from_string), save some time by not passing it through again at model.Molecule creation time.

Bug Fixes

  • (GH#48) Fixed a Molecule.get_fragment bug where ghosted fragments still asserted charge/multiplicity to the validator, which was rightly confused.

0.3.1 / 2019-03-07


  • (GH#37) Documentation now pulls from the custom QC Archive Sphinx Theme, but can fall back to the standard RTD theme. This allows all docs across QCA to appear consistent with each other.

  • (GH#41) Conda-build recipe removed to avoid possible confusion for everyone who isn’t a Conda-Forge recipe maintainer. Tests now rely on the conda env setups.

  • (GH#44) Molecule objects are now always validated against a more rigorous model and fragment multiplicities are fixed at the correct times, even when no multiplicities are provided. Molecule defaults to dtype=2.

Bug Fixes

  • (GH#39) Fixed to call pytest instead of unittest when running tests on install

  • (GH#41) Pinned a minimum Pytest version to make sure errors are not because of too old of a pytest version

0.3.0 / 2019-02-27

New Features

  • (GH#33) molparse.to_schema recognizes dtype=2 in keeping with GH:MolSSI/QCSchema#60 with internal schema_name=qcschema_molecule and schema_version=2 fields. molparse.from_schema recognizes external fields (existing functionality), internal fields (dtype=2), and mixed.

  • (GH#33) Pydantic molecule model now contains schema_name and schema_version=2 information.

  • (GH#35) Models now have an extra field for extra attributes, no additional base keys are allowed.


  • (GH#34) Converts qcel.Datum to Pydantic model. Changes: (a) comment, doi, glossary fields must be accessed by keyword, (b) to_dict() becomes dict() and instead of only label, units, data fields in dict, now comment, doi, glossary present _if_ non-default, (c) complex values no longer list-ified by to_dict().

  • (GH#36) Changelog and Models documentation.

Bug Fixes

0.2.6 / 2019-02-18

Bug Fixes

  • (GH#32) Updates compliance with Pydantic v0.20.

0.2.5 / 2019-02-13


  • (GH#31) Lints the code base preparing for a release and minor test improvements.

Bug Fixes

  • (GH#30) Fixes dihedral measurement code for incorrect phase in certain quadrants.

0.2.4 / 2019-02-08

New Features

  • (GH#27) Adds a new measure feature to Molecule for distances, angles, and dihedrals.

  • (GH#25) Adds a new testing module which contains testing routines for arrays, dictionaries, and molecules.


  • (GH#28) Reduces loading time from ~1 second to 200 ms by delaying pint import and ensuring git tags are only computed once.

0.2.3 / 2019-01-29


  • (GH#24) Update models to be compatible with QCFractal and MongoDB objects in the QCArchive Ecosystem. Also enhances the Molecule model’s json function to accept as_dict keyword, permitting a return as a dictionary of Pydantic-serialized python (primitive) objects, instead of a string.

0.2.2 / 2019-01-28

Bug Fixes

  • (GH#21) Molparse’s from_schema method now correctly parses the new qcschema_X strings for schema names.

  • (GH#23) Pydantic model serializations now correctly handle Numpy Array objects in nested BaseModels. Model serialization testing added to catch these in the future.

0.2.1 / 2019-01-27

  • (GH#20) Moves several Molecule parsing functions to the molparse module.

0.2.0 / 2019-01-25

  • now requires Python 3.6+

  • now requires Pydantic

New Features

  • (GH#14, GH#16, GH#17) Added new Pydantic models for Molecules, Results, and Optimizations to make common objects used in the QCArchive project all exist in one central, always imported module.


  • (GH#13) Function util.unnp that recursively list-ifies ndarray in a dict now handles lists and flattens.

0.1.3 / 2018-12-14

New Features

  • (GH#12) Adds “connectivity” validation and storage consistent with QCSchema.


  • (GH#12) Adds single dictionary provenance consistent with QCSchema rather than previous list o’dicts.

0.1.2 / 2018-11-3

New Features

  • (GH#10) Adds covalent radii data available through covalentradii.get(atom) function.

  • (GH#10) Adds to_units(unit) to Datum class to access the data in non-native units.

  • (GH#10) Adds periodictable.to_period(atom) and to_group(atom) functions to address periodic table.

0.1.1 / 2018-10-30

New Features

  • (GH#7, GH#9) Adds “comment” and “provenance” fields to internal repr to better match QCSchema.

  • (GH#7) Adds provenance stamp to from_string, from_arrays, from_schema functions.


  • (GH#7) Adds outer schema_name/schema_version to to_schema(..., dtype=1) output so is inverse to from_schema.

Bug Fixes

  • (GH#8) Tests pass for installed module now that comparison tests are xfail.

0.1.0a / 2018-10-24

This is the first alpha release of QCElemental containing the primary three components.

New Features

  • (GH#6) Updated molparse to write new Molecule QCSchema fields in keeping with GH:MolSSI/QCSchema#44

  • Periodic Table data from NIST SRD144 (c. pre-2015?) collected into qcelemental.periodictable instance, with accessors to_Z, to_element, to_E, to_mass, to_A (and redundant accessors to_mass_number, to_atomic_number, to_symbol, to_name) in float and Decimal formats. Also includes functionality to write a corresponding “C” header.

  • Physical Constants data from NIST SRD121 (CODATA 2014) collected into qcelemental.constants instance, with access through qcelemental.constants.Faraday_constant (exact capitalization; float result) or get (free capitalization; float or Decimal result). Also includes functionality to write a corresponding “C” header.

  • molparse submodule where from_string, from_array, from_schema constructors parse and rearrange (if necessary) and validate molecule topology inputs from the QC and EFP domains into a QCSchema-like data structure. Current deficiencies from QCSchema are non-contiguous fragments and “provenance” fields. Accessors to_string and to_schema are highly customizable.

  • A pint context has been built around the NIST physical constants data so that qcelemental.constants.conversion_factor(from_unit, to_unit) uses the QCElemental values in its conversions. Resulting float is within uncertainty range of NIST constants but won’t be exact for conversions involving multiple fundamental dimensions or wavelength -> energy != 1 / (energy -> wavelength).