atomref 0.1.4

Curated atomic reference data and transfer policies for geometry and structure-analysis algorithms.

atomref

Documentation: https://delonecommons.github.io/atomref

atomref is a small pure-Python package for curated atomic reference data and provenance-aware lookup policies used by geometry and structure-analysis algorithms.

It is not meant to be yet another periodic-table encyclopedia. The package is for code that needs stable atomic reference values with explicit provenance, clear fallback behavior, and honest handling of incomplete preferred datasets.

What you get in the current release line:

• stable element metadata,
• curated named radii sets,
• provisional X–H bond-length support for hydrogen-normalisation workflows,
• dataset provenance and coverage metadata,
• deterministic lookup policies,
• substitution and linear transfer from support datasets or policies into target datasets,
• guarded nested policy-backed transfers with explicit transfer depth, conservative fit/prediction controls, and cycle detection,
• user-defined custom element-indexed scalar sets.

Core terms

atomref uses a small vocabulary on purpose.

• quantity — the operational property family being requested, such as covalent_radius, van_der_waals_radius, atomic_radius, or xh_bond_length.
• domain — the key space used to index that quantity. In the current runtime, the supported domain is element, meaning lookups are keyed by an element symbol.
• dataset — one curated named table inside a quantity, such as cordero2008, alvarez2013, or csd_legacy_xh_cno.
• policy — the ordered rule set that decides what value to return when the preferred dataset is incomplete.

The metadata layer already records domain explicitly because the package is built for later extension, but the current runtime intentionally keeps the implementation narrow and stable: the current runtime resolves only element-domain scalar values.

Why this exists

Scientific software often wants a complete lookup table, but the best dataset for the job is rarely complete. atomref makes that situation explicit. Instead of hiding ad hoc defaults inside algorithm code, you choose a target set, describe how missing values may be restored, and keep provenance on what was actually returned.

The built-in default behavior is intentionally simple and practical:

• Cordero covalent radii (cordero2008) are the preferred covalent target set, with missing values substituted from the legacy CSD covalent radii (csd_legacy_cov).
• Alvarez van der Waals radii (alvarez2013) are the preferred vdW target set, with missing values restored from the Rahm isodensity atomic radii (rahm2016) through a fitted linear transfer.
• CSD/ConQuest hydrogen-normalisation defaults (csd_legacy_xh_cno) are a provisional sparse X–H target set for C, N, and O, with other parent elements inferred from Cordero covalent radii through a fitted linear transfer.

Nested policy predictors are supported too. LinearTransfer separates fit-time use of nested predictor values from prediction-time use. By default, the fit may use only direct nested values, while the final requested element may still use one additional nested completion step. That is a useful compromise for workflows such as provisional X–H inference from a chosen covalent-radii policy.

Quick example

>>> import atomref as ar
>>> ar.get_covalent_radius("C")
0.76
>>> ar.get_vdw_radius("O")
1.5
>>> ar.get_xh_bond_length("N")
1.015
>>> lookup = ar.lookup_vdw_radius("Pm")
>>> lookup.value
2.8972265395148358
>>> lookup.source
'transfer_linear'
>>> lookup.transfer_depth
1
>>> lookup.resolved_from
(DatasetRef(quantity='atomic_radius', set_id='rahm2016'),)

get_* returns only the number. lookup_* returns a LookupResult that also records where the value came from, whether a transfer model or policy source was involved, and how many transfer steps were needed (transfer_depth).

You can inspect the packaged quantity and dataset catalog directly:

>>> import atomref as ar
>>> ar.list_quantities()
('covalent_radius', 'van_der_waals_radius', 'atomic_radius', 'xh_bond_length')
>>> ar.get_quantity_info("xh_bond_length")
QuantityInfo(quantity='xh_bond_length', domain='element', units='angstrom', description='Element-indexed reference X-H bond lengths keyed by parent element X and intended for hydrogen-position normalisation or related geometry workflows.')
>>> [info.ref.set_id for info in ar.list_dataset_infos("van_der_waals_radius", usage_role="target")]
['bondi1964', 'rowland_taylor1996', 'alvarez2013', 'chernyshov2020']

You can also load a packaged set directly:

>>> import atomref as ar
>>> vdw = ar.get_radii_set("van_der_waals", "alvarez2013")
>>> vdw.get("O")
1.5
>>> xh = ar.get_xh_set("csd_legacy_xh_cno")
>>> xh.get("C")
1.089

Notebook walkthroughs

The repository ships example notebooks for the main workflows. In the documentation they are also available as rendered Markdown pages, so users can read them without opening Jupyter first.

• Notebook overview
• Quickstart notebook
• Policies and assessment notebook
• Custom sets and discovery notebook

Relationship to Delone Commons

atomref is designed as a standalone package, but within Delone Commons it is primarily intended to support chemistry-aware packages such as:

• molcryst, for covalent-bond detection, contact analysis, and hydrogen workflows,
• future chemvoro, for chemistry-aware contact and hydrogen workflows.

By contrast, pyvoro2 and pbcgraph are intentionally general mathematical packages and are not direct consumers of atomref.

Data curation and developer tools

The repository also ships small maintenance tools. The most important ones are:

• python tools/check_registry.py — validate curated registry metadata against packaged CSV tables,
• python tools/check_notebooks.py — execute notebook code cells,
• python tools/export_notebooks.py — turn notebooks into Markdown pages for the docs,
• python tools/gen_readme.py — regenerate README.md from this page,
• python tools/release_check.py — run the full release-preparation checklist, including linting, tests, docs, builds, and artifact validation.

See the tools README for a short description of each script.

---

This README is generated from docs/index.md.

To regenerate it:

python tools/gen_readme.py

Edit the documentation sources instead of editing README.md directly.