Changelog

All notable changes are recorded here, in the style of Keep a Changelog. The review register (REGISTER.md) records the changes to data and models in detail. Versions before 0.8.0 were internal and were not published.

Unreleased

No recommended value or model changed.

Added

• Helium on matched measures (TC-1092): a new source of tier-3 calculations, travers_2026_pyroraptor (both electrons active, haCC with one and five ion channels, and one electron, at 390 nm; haCC in Sato et al.’s 800 nm pulse), and gas.ionisation_validation.helium_matched_ratios.

• Xenon in Hartmann et al.’s RMT field (TC-1093, TC-1094): one 5p electron in the same field and with the same measure (travers_2026_trecx, part N), the paper’s measured neutral abundance (hartmann_2022), and xenon_rmt_ratios and xenon_survival.

• The curated data are now 762 source records: 8481 datasets and 197,804 values, 259 errata and 780 published formulas.

• gas.ionisation_rates.ionisation_rate_coverage and gas.molecular_ionisation_rates.molecular_rate_coverage: the gases each rate model covers, read from the rates file without building a model, so a caller can tell an atom from a molecule.

• gas.models.fitted.fitted_model(name) and gas.models.formulas.published_model(name): one gas model, reading only its own record.

• Source.dataset(id) and load_source(path, defer=True): a source’s datasets are built, and their data files read, when first asked for.

• A catalogue index, data/catalogue_index.json, written by tools/catalogue_index.py: which record holds the datasets and formulas of each material and quantity. It is checked against the records on first use; an out-of-date index raises the new CatalogueIndexWarning and the records are read instead.

Changed

• The models read only the data they need (no value changed). A model that needed a few datasets read the whole catalogue (8,300 files), and a gas’s model built every gas’s: molecular_rate_model("N2") read 8,671 files and now reads 35, ionisation_rate_model("He") 339 and now 34, gas default_model("He") 195 and now 28, air_model() 799 and now 19, and the SiO2 models 3,964 and now 92. Datasets are looked up by key, each gas’s model, rate and parameters are built on demand, searches by material, quantity or formula use the index, and a source’s data files are read only for the datasets used. Cold starts, where every file opened is read from disk or scanned, are seconds to tens of seconds shorter.

• Source.datasets is a cached property built from the new Source.entries (the record’s dataset tables), no longer a constructor field.

Fixed

• The manual’s helium findings at 390 nm (TC-1092 corrects TC-1050): haCC’s helium is 0.96 to 0.99 of the exact atom, not 0.76 to 0.81, and the atom 1.6 times one electron, not 1.7 to 1.8. The earlier ratios compared a norm-loss that counts Rydberg excitation with one that does not. Parker et al.’s rate is confirmed as before.

0.8.1 (2026-10-16)

Fixes for a downstream propagation code. No value at a positive wavelength, model or stored fit changed.

Added

• optmat’s DOI, 10.5281/zenodo.23422879 (Zenodo; every version), in CITATION.cff and the README.

• A py.typed marker (PEP 561), so that type checkers of downstream code read optmat’s annotations.

• optmat.solid.models.formulas.published_model(name): one published formula, reading only its source’s record.

Changed

• A model refuses a wavelength at or below zero, and a solid model refuses inf, with ArgumentError; before, a negative wavelength returned a finite, meaningless value (TC-1091). A gas and air keep inf as the static limit; NaN still gives NaN.

• Faster lookups for a propagation code. solid.models.registry.model (and so default_model) builds a published formula from its own record, not from every record with formulas (N-BK7, BaF2, MgF2: 0.3 s from about 2 s in a fresh process), and gas.air.air_model reads only the records with mole fractions, not the whole catalogue (about 1.4 s from 4-5 s). No value changes.

0.8.0 (2026-10-16)

The first public release, at github.com/LupoLab/optmat. No value, model or stored fit changed.

Changed

• Wider support: Python 3.12 and later (was 3.14) and NumPy 2.3 (was 2.5); SciPy stays at 1.18. These are the oldest versions that reproduce the stored results: Python 3.12 for PEP 695 syntax; NumPy 2.3 and SciPy 1.18 because the stored fits and the models’ identity digests, which hash computed arrays, reproduce bit for bit only from these releases. CI tests them.

• The gas extra no longer installs pyfluids, which optmat never imported; it calls CoolProp directly.

• Coverage is measured with pytest-cov (as in solprop), replacing tools/measure_coverage.py; the gate still requires 100 % statement and branch coverage, including ordinary-Python subprocesses, and writes an XML report for CI.

• The register’s branch-integration recipe moved from AGENTS.md to Maintaining the review register. Agents create their topic branch and worktree without asking and keep the main checkout clean.

• Private development references (task briefs, machine paths) removed from the tRecX archive, the register and the source records; wording only.

Added

• CI (.github/workflows/ci.yml): the full gate on Linux and macOS, and the tests on Python 3.12 with the lowest supported dependencies; a release workflow publishing to PyPI by trusted publishing.

• Every file states its copyright and licence (REUSE): REUSE.toml, LICENSES/, SPDX headers in the code, and reuse lint in the gate. Code, source records and fits are MIT, prose and figures CC-BY-4.0. The published data the package ships (the values transcribed into data/values/ and the files in data/external/) remain their publishers’, included under a third-party data notice (LICENSES/LicenseRef-optmat-third-party.txt) so that the package is self-contained; each external directory’s PROVENANCE.md states its publisher’s terms.

• CITATION.cff and .zenodo.json.

• Windows: every text file is read and written as UTF-8, whatever the platform’s default (Windows decoded the curated records with its code page, which failed on their author names and unit signs); tests/test_encoding.py keeps it so, and CI runs the tests on Windows. A git checkout keeps LF line endings on every platform (.gitattributes), so the stored configurations’ hashes match there too; the archives under data/external/ are checked out byte for byte.

Fixed

• Citation metadata: Fernley, Taylor and Seaton’s page (6457, as reproduced in Berkowitz 1997), the co-authors of van der Hart et al. 2007, the pages of van Duijnen and Swart 1998 and the first page of Manakov et al. 1986. A source note no longer quotes a reference that does not exist.

• The manual: Blow and Wood’s α is their Kerr weight; Ammosov, Delone and Krainov define n* in their eq. 2; and the static rate is their eq. (1) without its cycle-averaging factor.

0.7.0 (2026-10-15)

Crystalline silicon’s complex refractive index from 31 nm to 143 µm at 20 °C: fit/Si, Franta et al.’s universal dispersion model refitted to Frey et al.’s absolute indices, Schinke et al.’s absorption edge and Franta et al.’s 20 °C tables, now the recommended model of Si (it had none), with twelve new source records and the manual page Crystalline silicon. Also argon’s rounds for OEP 8, RMT’s xenon ground-state survival, a five-section manual with a generated source catalogue, and refit tests that compare the stored solid, gas and compact fits on their predictions and objective rather than their parameters. No other recommended value changed; no stored fit changed.

Added

• A five-section manual: Introduction, User guide, The data, Developers guide, and API. The introduction includes installation guidance, a functionality overview, the quick start, and an explanation of curation; the user guide groups the scientific chapters in reading order, with previous/next navigation.

• A reproducible source catalogue generated during documentation builds from every source record, including citations, datasets, formulas, notes and corrections.

• Separate chapters for nonlinear conventions, modelling, support and extrapolation; Raman validation; atomic and molecular ionisation validation; and two-photon source coverage and bound excitation. Existing page URLs and moved section fragments lead to the new locations.

• Dedicated developer instructions for extraction, fitting, documentation and register maintenance. Historical solprop comparisons now live in regression fixtures; the public validation retains factual comparisons with an identified Luna.jl snapshot. The manual, gallery, README and API descriptions use consistent terminology and distinguish current defaults from historical validation results.

• The vocabulary quantity ground_state_survival: the probability that a pulse leaves the atom in its ground state, read at a stated time (TC-1058).

• hartmann_2022: RMT’s xenon ground-state survival at 759 nm, 1.1 to 2.5e14 W/cm2, from the authors’ archive, with the pulse read from the fields RMT wrote (a 1/e half-width of 186 au: the paper’s “FWHM 4.5 fs” is a 5.30 fs intensity FWHM). hutcheson_2025 is recorded without values: its archived populations were computed at about 670 nm, not the stated 760 nm (TC-1059).

• Argon’s round for OEP 8 (TC-1060 to TC-1068): eight sources.

  • The Ar 4p levels’ lifetimes and quenching by 22 gases after two-photon excitation at 184-189 nm (Sadeghi et al. 2001), confirming Wiese et al.’s radiative widths.

  • Measured and calculated Ar 4p-4s transition probabilities (Asghar et al. 2020; Savukov 2018).

  • Calculated photoionisation of every Ar, Kr and Xe np level near threshold (Petrov et al. 2011).

  • Kr and Xe lifetimes and transition probabilities (Aymar and Coulombe 1978).

  • Three Stark-effect papers, as notes.

• Argon’s records for OEP 8 (TC-1069 to TC-1074):

  • Chang and Setser’s Ar 4p lifetimes, quenching and transition probabilities, and Nguyen and Sadeghi’s state-to-state transfer (both 1978). Together they show that Sadeghi et al.’s self-quenching of 4p[1/2]0 and 4p[5/2]2 is low by about 2.

  • Baig et al.’s measured 4p’ photoionisation at threshold (2008), 2.8-3.1 times Petrov et al.’s calculation.

  • One-photon 4s-4p pressure broadening and shifts in Ar (Sadeghi et al. 2022) and Ar-He (Eshel et al. 2016).

  The curated data are now 749 source records: 8357 datasets and 166,428 values, 255 errata and 771 published formulas.

• Crystalline silicon’s complex index (TC-1075 to TC-1088): fit/Si, the recommended model of Si from 31 nm to 143 µm at 20 °C (manual page Crystalline silicon).

  • Franta et al.’s universal dispersion model of float-zone silicon (the newAD2 parameter file, now franta_2017/udm_c_si, 0.016 high in n in the near infrared and 2.6 times high in k at 633 nm), refitted to Frey, Leviton & Madison’s absolute indices at 1.2-5.5 µm (taken from 295 K with their dn/dT; within 1.4e-4), to Schinke et al.’s absorption coefficient at 250-1200 nm, and to Franta et al.’s temperature-dependent model at 20 °C (their refractiveindex.info tables) in the ultraviolet and the infrared lattice bands.

  • Twelve source records: Frey et al. 2006 (Si and Ge, 30-295 K, with their dn/dT and temperature-dependent Sellmeier formulas), Li 1980, Green 2008, Schinke et al. 2015, Aspnes & Studna 1983 (Si, Ge, GaAs, GaSb, InP), Jellison 1992, Herzinger et al. 1998, Edwards’s chapter in Palik’s handbook (1985), Philipp & Ehrenreich 1963, and three Franta et al. papers (2014, 2018, 2019); franta_2017 gains the newAD2 model and tables and the 20 °C, 25 °C and 300 K tables.

  • optmat.solid.fitting fits an absorption_coefficient as k = alpha lambda / (4 pi).

  • The model’s k beyond 1.15 µm is an upper bound, not the band-to-band absorption; it is for intrinsic float-zone silicon at 20 °C only. The curated data are now 761 source records: 8467 datasets and 197,754 values, 259 errata and 780 published formulas.

Fixed

• The solid fits’ refit tests (fit/SiO2, fit/Si) compare parameters in units of their stored uncertainties, predictions over the model’s range and the objective, instead of each parameter to 1e-6 relative. Silica’s stored fit had never matched a refit at that precision: the optimiser stops at a path-dependent point of a flat minimum, 0.003 sigma away, at the same chi-squared to 2e-9 (TC-1089). No stored value changed.

• The gas dispersion fits’ and compact models’ refit tests compare predictions inside each fit’s intervals (to 5 % of the model’s band and 1e-4 relative), the full objective (to 0.01 in chi-squared) and the compact models’ residuals (to 0.01 sigma), instead of the polarizability to 1e-7 out to 65 nm and the poles to 1e-9. Every stored gas fit and compact model had stopped elsewhere on a nearly degenerate minimum than a refit of the committed code does, with the same data objective to round-off (TC-1090). No stored value changed.

• Numerical regression checks tolerate demonstrated arithmetic differences without changing model evaluation or stored fits. Gas Raman golden tests compare canonical identity records as well as exact self-consistent hashes; the existing golden numerical values are preserved. Nonlinear refit tests check predictions, parameter displacement in uncertainty units, dataset scales and the full objective including priors. Mutation tests retain sensitivity to substantive regressions (OEP 1, Q32).

• Review of the manual’s restructure:

  • The Raman page no longer assigns CH4 and SF6 to the normal-coordinate strengths (they use Schrötter & Klöckner’s Σ).

  • D2’s Q-width ratios are stated with their density dependence.

  • The measured-yield comparison gives the current default’s intensity scales and ratios, not the static-factor default’s.

  • An open question of the ionisation accuracy page and the ratio direction of the static-rate table are restored. So are the capillary zero-dispersion sensitivity and the bound on the temperature dependence of the index.

  • The CH4 gallery caption’s plotted range and the Zhang eq. (8) citation are corrected. Cross-references left behind by the page split are retargeted.

  • The moved solprop tables have their column definitions back.

0.6.0 (2026-10-12)

The bound two-photon levels of the noble gases and the literature behind OEP 8: optmat.gas.two_photon_bound, the even levels below threshold from NIST with every curated two-photon strength in one per-level convention and the excited levels’ photoionisation (OEP 6, Phase 2a; OEP 6 is Final). Seven vocabulary quantities for radiative, collisional, light-shift and amplitude data. About 160 sources on two-photon dynamics, transition probabilities, lifetimes, quenching, pressure effects, hydrogen’s exact amplitudes and the xenon n2. The maintainer’s two-photon tRecX runs, archived, confirm the helium model and OEP 8’s weak-field form. Also the strong-field dissociation data of OEP 7 (Draft) and helium with two active electrons at 390 nm. No recommended value changed. The default ionisation-rate model’s helium digest changed with its source record travers_2026_trecx (TC-1050); its values are unchanged.

Added

• OEP 6 is Final (OEP 6, Two-photon ionisation and absorption of the noble gases): the continuum model optmat.gas.two_photon and the bound-level records optmat.gas.two_photon_bound. Its Phase 2b and Phase 3 pass to OEP 8.

• travers_2026_trecx: helium with two active electrons at 390 nm, haCC’s helium and one electron in the same pulse; parker_pulse_yield, helium_two_electron_ratios. The atom is 1.7 to 1.8 times one electron; Parker et al.’s rate over the pulse is within 1 to 5 % of it; haCC’s helium is 0.76 to 0.81 of it. The manual’s helium table labelled the static factor’s column default since 0.3.0; corrected (TC-1050). The curated data are now 734 source records: 8162 datasets and 164,315 values, 255 errata and 771 published formulas.

• NIST’s near-infrared lines of Ne, Ar, Kr and Xe, held as returned beside the resonance lines (TC-983): the transition probabilities among excited levels, for two-photon amplitudes, in vacuum wavelengths throughout (re-queried, TC-985).

• The literature of OEP 8 (TC-986 to TC-1010): 24 sources on two-photon dynamics, transition probabilities, lifetimes and the xenon n2 near 248 nm. Taken as data: three Kr level energies and the ionisation limit of the 86Kr isotope (Brandi et al. 2002), the Kr and Xe resonance oscillator strengths (Matthias et al. 1977), and reproduced He and Xe alpha and gamma (Manakov et al. 1986). The A values, lifetimes, quenching rates and excited-level polarisabilities await vocabulary.

• Five quantities for the bound levels’ radiative and collisional data (TC-1011): transition_probability, radiative_lifetime, level_polarizability, light_shift_coefficient and collisional_quenching_rate_coefficient. Their data (TC-1012 to TC-1020):

  • the A values of Ar 4s-4p and 4s-5p (Wiese et al. 1989), Kr 5s-5p (Dzierżęga et al. 2000) and Xe 6s-6p (Sabbagh and Sadeghi 1977);

  • the lifetimes of the Xe 6p, 6p’ and 7p levels (Bruce et al. 1990), the Kr and Xe resonance levels (Matthias et al. 1977) and the Kr 5p levels;

  • Xe quenching by Xe and Ar;

  • the static polarisabilities of the Ar 4s and 4p levels (Werbowy 2026);

  • the Kr 5p[1/2]0 light shift (Brandi et al. 2002).

• The second round of OEP 8’s literature (TC-1021 to TC-1041): 22 sources.

  • Two quantities (TC-1040): pressure_line_coefficient and two_photon_transition_amplitude (bound levels). Lifetimes, quenching and excited-level photoionisation now also apply to molecules.

  • Hydrogen’s exact two-photon amplitudes against the photons’ split (Quattropani et al. 1982, Drachman et al. 1990).

  • Xe 6p lifetimes and quenching (Bowering et al. 1986, Ku and Setser 1986), pressure shifts and broadening (Gornik et al. 1981, Hawkins et al. 1980).

  • J-averaged excited-state photoionisation of Ne, Ar, Kr and Xe (Duzy and Hyman 1980).

  • The H2 and D2 dissociation energies and the Kr 5p[1/2]0 light shift (Balakrishnan et al. 1992, 1994).

• The third round of OEP 8’s literature (TC-1042 to TC-1047): the Xe 6p’, 7p and 6d lifetimes and quenching by Ne and Ar (Inoue et al. 1984), which put the 6p’ lifetimes well above NIST’s; the H2 E,F lifetime and quenching from their original (Kligler and Rhodes 1978); the hydrogen 1s-2s amplitude in both gauges (Bassani et al. 1977); He 1s2p and 1s3p lifetimes and photoionisation (Johansson et al. 2003); two-photon Rabi oscillations in excited He (Fushitani et al. 2015).

• The maintainer’s tRecX runs of two-photon ionisation and excitation (TC-1048, TC-1049), archived with their inputs: the two-photon ionisation cross sections of He (two-electron haCC and one-electron) and Ar (one-electron), and the one-electron Ar 3p-4p amplitudes, as tier-3 checks. haCC’s helium agrees with optmat’s model to 0.90-1.16. The two-colour runs support OEP 8’s pair-form rule for the He and Ar continua. Run 5b (TC-1051) puts haCC’s propagated He 1s2s at 20.739 eV: with it, the 1s2s populations follow OEP 8’s weak-field form to 1 %, and the fitted |M|^2 = 12.31 au^2 equals Hattig et al.’s strength to 0.04 %. Run 5c (TC-1052) finds haCC’s propagated 1s2p where its state list puts it (21.2089 eV), so run 2’s pole distances stand; haCC’s 1s2p dipole, 2 % below NIST’s, is held as a check.

  The curated data are now 734 source records: 8151 datasets and 164,296 values, 255 errata and 771 published formulas.

• Strong-field dissociation data for O2, N2, N2O, NO2, O3 and N2O5 (TC-954 to TC-981): eight quantities (dissociation energies, ionic-state populations, fragmentation branching ratios, kinetic energy release, dissociation times, dissociative recombination coefficients, photodissociation quantum yields and transition dipoles); NO2, N2O5, NO, NO2+, NO+, N2++ and O2++ registered by reference only; 27 sources curated, among them the density-matrix tables of Yuen and Lin (2022, 2023) and Jia et al. (2024), Kleine et al.’s measured N2+ state populations (2022), Berkowitz and Eland’s N2O+ branching ratios (1977), and Kossyi et al.’s recombination coefficients (1992). No dissociation model yet. The curated data are now 682 source records: 7947 datasets and 163,579 values, 232 errata and 771 published formulas.

• The bound two-photon levels of the noble gases (OEP 6, Phase 2a; TC-952, TC-953): optmat.gas.two_photon_bound.two_photon_bound_levels(gas). It gives every even J = 0, 1, 2 level below the ionisation energy from NIST, with its curated two-photon strengths in one convention (per level; sigma0_from_delta), measured and calculated, and its photoionisation cross sections. Helium and xenon are checked, neon, argon and krypton unresolved. There is no evaluator (Phase 2b waits on OEP 6 Q9).

• Five further two-photon sources requested for krypton (TC-946 to TC-950): Ritchie 1977 and L’Huillier, Jönsson and Wendin 1986 (helium two-photon ionisation), Kulander 1987 (helium, time-dependent Hartree-Fock) and 1988 (xenon), and Crance and Aymar 1985 (a note-only record). None holds krypton data: krypton still has no two-photon model (OEP 6, Q2). OEP 6 Phase 2 is restricted to curation for now (TC-945). The curated data are now 656 source records: 7810 datasets and 162,996 values, 232 errata and 771 published formulas.

Changed

• hummer_1998’s threshold cross sections are placed at the calculation’s own thresholds (TC-951, correcting TC-934). gangopadhyay_1986’s three-photon table carries a doubt: Kulander 1988 reports a conversion-factor error of about 1e6 (TC-949).

0.5.0 (2026-10-09)

Two-photon ionisation of the noble gases: optmat.gas.two_photon, the generalised two-photon ionisation cross sections of He, Ne, Ar and Xe (Xe also circular), each the reference curve the author approved, with thresholds, labelled regions (NaN where unavailable, never an error on a grid), uncertainty by kind and an identity (OEP 6, Phase 1; TC-943, TC-944). The two-photon literature curated on the way: four vocabulary quantities and 66 sources on two-photon ionisation, bound two-photon excitation and excited-level photoionisation, compared on one footing in the manual (TC-872 to TC-942). Krypton has no model; no existing value or digest changed.

Added

• Two-photon ionisation of the noble gases (OEP 6, Phase 1; TC-943, TC-944): optmat.gas.two_photon.two_photon_ionisation_model(gas, polarization), the generalised two-photon ionisation cross section of He, Ne, Ar and Xe (Xe also circular). Each model is the reference curve the author approved, with its two-photon thresholds, E1, labelled regions (zero below threshold; NaN where unavailable, never an error on a grid), its uncertainty kept by kind, rate, loss_coefficient, pair_cross_section and an identity. Krypton has no model. New exceptions ArgumentError and UnavailableModelError. No existing model, default or digest changes.

• The follow-up two-photon literature (TC-904 to TC-942): the originals the first round quoted, and the other calculations it named. Among them are two-photon ionisation of He (van der Hart and Bingham 2005; Bachau et al. 1990; Proulx and Shakeshaft 1993), Ne and Ar (Moccia et al. 1983; Pan, Gao and Starace 1990; Pindzola and Kelly 1975), Ar, Kr and Xe (Fink and Johnson 1990, relativistic, linear and circular) and Xe (Gangopadhyay et al. 1986; L’Huillier and Wendin 1987; Bischel et al. 1979; Hodges et al. 1981). There are also bound two-photon measurements and calculations (McCann et al. 1988; Khambatta et al. 1988; Bokor et al. 1980; Sundholm et al. 1994; and others), excited-level photoionisation (Chang and Kim 1982; Pindzola 1981; Hyman 1977; Hummer and Storey 1998) and xenon’s Rydberg levels (Yoshino and Freeman 1985). Krypton still has no vector-drawn or tabulated two-photon ionisation cross section (OEP 6, Q2). The manual page’s comparison is extended. OEP 6’s author decided that the model will not raise on propagation grids, and that a time-resolved loss current is a documented gap (TC-903). The curated data are now 651 source records: 7785 datasets and 162,533 values, 232 errata and 771 published formulas.

• Four quantities for two-photon data (TC-872): two_photon_transition_strength, two_photon_excitation_cross_section, excited_state_photoionization_cross_section and multiphoton_ionization_polarization_ratio.

• The two-photon literature of the noble gases (TC-873 to TC-901): the two-photon ionisation cross sections of He (Saenz and Lambropoulos 1999; Mihelič and Horvat 2021; Nikolopoulos and Lambropoulos 2001; Sato et al. 2011, measured), Ar (Petrov et al. 2016; Starace and Jiang 1987), Ne to Xe (McGuire 1981, circular), Kr and Xe (L’Huillier, Tang and Lambropoulos 1989; L’Huillier and Wendin 1987) and Xe (Lagutin et al. 2017, linear and circular; Karamatskou and Santra 2017; McCown et al. 1982, measured; Santra and Greene 2004); bound two-photon strengths and cross sections (Hättig et al. 1998; Henriksson et al. 2005; Shekhtman et al. 2020; Kröll and Bischel 1990; Drag et al. 2021; Dakka et al. 2018) and excited-level photoionisation (with Geohegan et al. 1986); and records of the papers that hold nothing to curate. The curated data are now 613 source records: 7641 datasets and 155,228 values, 226 errata and 771 published formulas.

• A manual page, Two-photon ionisation and absorption: the data, comparing them on one footing, with the thresholds of each gas (TC-902). optmat has no two-photon model yet; that is OEP 6’s.

Changed

• The analysis tool behind the third-order manual’s tables, tools/fit/nonlinear_support.py, reads each gas’s resonance limits from its support record, so data/gas/nonlinear_support.toml is the one copy of the selections (OEP 5, Q11; TC-871). No number in the tables changed.

0.4.0 (2026-10-07)

Where the hyperpolarizability fits are supported: each model’s support, a public record of the largest fitted W2, single leg and pair sum and of the gas’s resonance limits, with their sources, for propagation codes that evaluate every four-wave quartet leg by leg (OEP 5, Final; TC-862 to TC-870). New sources on ionisation and the third order: many-electron calculations against one electron, R-matrix multiphoton cross sections, ion yields, saturation intensities, xenon’s negative n2 and helium’s dynamic polarizability (TC-852 to TC-861).

Added

• OEP 5 is Final (OEP 5, A public record of where the hyperpolarizability fits are supported): HyperpolarizabilityModel.support and its records (TC-870).

• HyperpolarizabilityModel.support, a public record of where each gas’s hyperpolarizability fit is supported, for propagation codes (OEP 5, Phase 1; TC-870). It holds the largest fitted W2, single leg and pair sum, with the fitted points that attain them, and the gas’s resonance limits E1, E2 for one linear polarisation and for any, and the ionisation energy, each with its datasets, kind, qualifications and uncertainty; absorption evidence is marked as such and a missing limit has its reason. It also gives two derived quantities, leg_max_in_w2 and band_edge(). Wavenumbers are in m^-1. The record states facts and carries no rule. It has its own digest and leaves every model’s digest and evaluated gamma unchanged. New classes HyperpolarizabilitySupport, SupportLimit, SupportPoint, SupportUncertainty, SupportCandidate and AbsorptionEvidence; the selection data/gas/nonlinear_support.toml, the stored records data/gas/models/nonlinear_support/ and the tool tools/fit/nonlinear_support_record.py.

• Where the hyperpolarizability fits are supported, for propagation codes that evaluate every four-wave quartet leg by leg (solprop’s SEP 6): the legs of every fitted value and validator (TC-864); the resonance limits of the 15 gases, a single leg at the lowest dipole-allowed level and a pair sum at the lowest two-photon level (TC-865); the fits’ increase with each squared leg (TC-866); per-leg caps against the data, a Kerr carrier and solprop’s demonstration quartets (TC-867); and the quantities derived from the facts, sqrt(3 W2,max / 4) and min(E1, E2/2), with their use left to the propagation code (TC-868). New manual sections on Third-order nonlinearity and Third-order accuracy, and the tool tools/fit/nonlinear_support.py. The noble gases’ NIST ASD levels (TC-862) and Huber & Herzberg’s states of H2, D2, N2, O2 and CO from the NIST WebBook (TC-863), with a new quantity, electronic_state_energy. The curated data are now 585 source records: 7441 datasets and 113,554 values, 209 errata and 771 published formulas.

• travers_2026_trecx: haCC’s argon in a 20 fs pulse and with mMax 1, and one electron at its ionisation energy in both pulses: the atom about 0.8 of one electron, rising by 3 to 8 % from 6 to 20 fs (TC-861). The curated data are now 585 source records: 7402 datasets and 113,515 values, 209 errata and 771 published formulas.

• pathak_2020, pathak_2022, sato_2016, sato_2018: time-dependent many-electron calculations of Ar, Kr and He in short sin^2 pulses, read from the figures (TC-859); travers_2026_trecx: one electron in the same pulses (TC-858); and published_atom_ratios: argon 0.87 to 1.10 and krypton 0.91 of one electron, helium 1.53 to 1.71 (TC-860). The curated data are now 585 source records: 7392 datasets and 113,505 values, 209 errata and 771 published formulas.

• mckenna_2004: McKenna & van der Hart’s R-matrix Floquet two- and three-photon cross sections of Ne and Ar, read from the vector figures. Argon’s at 193 nm is matched within 26 % by the default with the sum over channels, and 1.5 to 12 times over with its integral, which gives the multiphoton yield a non-integer order (TC-856).

• lai_2017: Lai et al.’s ion yields of Ne, Ar, Kr and Xe from 0.4 to 3.9 um, linear and circular, read from the vector figures; their intensities partly fitted to PPT, so a test of the yields’ shape (TC-857). The curated data are now 581 source records: 7358 datasets and 113,427 values, 209 errata and 771 published formulas.

• uiterwaal_1998: the saturation intensities of He, Ar and Kr at 248.6 nm, 500 fs, from the ponderomotive shift, and their multiphoton cross sections; the quantity saturation_intensity; saturation_test. In this multiphoton regime (Keldysh parameter 1.8 to 7) the default’s rate is 0.55 to 4.3 times what they imply, the static Coulomb factor’s 5.7 to 32 (TC-853).

• lehmberg_1995: xenon’s negative n2 1 nm above its two-photon resonance at 248.4 nm, measured and calculated, and CO2’s (TC-854).

• masili_2003: helium’s dynamic polarizability to 0.845 a.u., through the 1s2p resonance; the library’s lost minus signs restored (TC-855). The curated data are now 579 source records: 7335 datasets and 102,921 values, 209 errata and 771 published formulas.

• tolstikhina_2014: the Hartree-Fock weak-field coefficients of He, Ne, Ar, Kr and Xe (Table I), and weak_field_factors, the atom’s weak-field rate over the model’s one electron: 1.7 to 1.8 for He, Ar, Kr and Xe, the literature’s factor 2 less the excess of Tong, Zhao & Lin’s coefficients; 0.53 for neon, unexplained (TC-852). The curated data are now 576 source records: 7322 datasets and 102,878 values, 209 errata and 771 published formulas.

Fixed

• Third-order nonlinearity said a calculated THG enters the fits with harmonics at 217 nm and longer; the 650 nm window keeps 694.3 nm and longer, harmonics at 231.4 nm (TC-864). The fits are unchanged.

• Third-order accuracy’s table of solprop’s demonstration quartets showed another table’s rows (TC-868).

• The support analysis is qualified after OEP 5’s review: the band edge min(E1, E2/2) is a sufficient bound, an SPM resonance only for one polarisation; the calculations beyond the data prove no general lower bound; the polyatomic molecules’ absorption numbers are evidence, not levels (TC-869).

0.3.1 (2026-10-06)

Xenon in Wallace et al.’s pulse (TC-849) and xenon’s effective number of electrons 1.5 (TC-850); krypton’s 1.35 rests on Wallace et al.’s yields, Santra et al.’s rate being an independent-electron sum, and the effective number of electrons is a calibration of the formula, not a count of active electrons (TC-851).

Fixed

• Santra et al.’s (2006) krypton rate is an independent-electron sum in a Herman-Skillman potential, not an all-electron calculation, as 0.3.0’s notes and krypton’s reasons said; krypton’s 1.35 electrons rest on Wallace et al.’s yields. The manual states that the effective number of electrons calibrates the formula and is not a count of active electrons (TC-851).

Changed

• Xenon’s effective number of electrons in the default ionisation rate is 1.5 (was 1, provisional), from Wallace et al.’s yields through the one-electron calculation in their pulse; its uncertainty factor is 3, for the other absolute measurements, which put xenon high against argon at 800 nm (TC-850).

Added

• travers_2026_trecx: one electron of xenon in Wallace et al.’s pulse at phases 0 and pi/2. Through their test it matches at 1.019, within the calibration, as argon’s does: the xenon atom ionises like about 1.1 electrons without the spin-orbit splitting, about 1.5 in the model’s terms. Xenon’s effective number of electrons stays 1, provisionally (TC-849). The curated data are now 575 source records: 7302 datasets and 102,858 values, 209 errata and 771 published formulas.

0.3.0 (2026-10-04)

The noble gases’ default ionisation rate takes the frequency-dependent Coulomb factor, validated against one-electron calculations of H, He, Ne, Ar, Kr and Xe, and krypton’s effective number of electrons rises to 1.35 (TC-848). Argon’s rate at 1e14 W/cm² and 800 nm falls by 24 %; the former default is kept as default_static. New one-electron and all-electron calculations from tRecX and haCC (TC-842 to TC-847).

Changed

• The default ionisation rate of the noble gases takes the frequency-dependent Coulomb factor (coulomb = "frequency", c = 5.25; TC-831), as does one_electron; krypton’s effective number of electrons is 1.35 (Santra et al.’s all-electron rate, Wallace et al.’s yields); xenon’s stays 1, provisionally. The former models are default_static and one_electron_static; the provisional frequency and one_electron_frequency are retired. Argon’s rate at 1e14 W/cm2 and 800 nm falls by 24 %. The manual explains the rate as three parts, each tested against what isolates it (TC-848).

• The table of van der Hart’s all-electron rates at 390 nm gains the candidate one_electron_frequency: with it one electron of argon is about 1.15 times the all-electron rates below gamma = 2.2, the atom at about 0.87 of one electron, as haCC’s interim all-electron calculation at 800 nm (TC-842).

Added

• travers_2026_trecx: the ionised fraction of atomic hydrogen at 400 and 800 nm in a 15 fs pulse, converged to 0.06 %, among the hydrogen comparisons: at 800 nm Zhao et al.’s, with the same pulse, are 0.96 times it on average (TC-843). The curated data are now 575 source records: 7280 datasets and 102,717 values, 209 errata and 771 published formulas.

• travers_2026_trecx: argon’s one-electron ionised fraction at 400 and 1030 nm with a 30 fs pulse, not fitted by the candidate frequency factor, which is 0.87 and 0.95 times it on average (the static factor 2.0 and 1.18) (TC-844). The curated data are now 575 source records: 7284 datasets and 102,752 values, 209 errata and 771 published formulas.

• travers_2026_trecx: one electron of krypton and xenon in A2’s 6 fs pulse and their static rates: the candidate frequency factor is 1.16 (Kr) and 0.94 (Xe) times the calculations, krypton’s excess its static normalisation (TC-845). The curated data are now 575 source records: 7288 datasets and 102,782 values, 209 errata and 771 published formulas.

• travers_2026_trecx: one electron of krypton in Wallace et al.’s pulse at phases 0 and pi/2; through their test it needs the intensities scaled by 1.045, beyond the calibration, where argon’s matched: the krypton atom ionises like about 1.35 electrons, as Santra et al.’s all-electron static rate also says (TC-846). The curated data are now 575 source records: 7290 datasets and 102,806 values, 209 errata and 771 published formulas.

• travers_2026_trecx: haCC’s all-electron argon in A2’s 6 fs pulse and one electron at haCC’s ionisation energies; hacc_ratios in optmat.gas.ionisation_validation: the argon atom ionises like 0.76 to 0.87 of one m = 0 electron at the same ionisation energy (TC-847). The curated data are now 575 source records: 7300 datasets and 102,834 values, 209 errata and 771 published formulas.

0.2.2 (2026-10-02)

solprop SEP 3’s findings: the ionisation rates three to four and a half times faster with the same numbers (TC-837); a smooth derivative of the solids’ indices made absolute (TC-838); D2’s viscosity from its correlation (TC-839); the Raman response’s status without the response (TC-840); and fused silica’s dispersion near 2 µm explained (TC-841). Also hydrogen’s one-electron calculations (TC-833 to TC-835) and argon’s in Wallace et al.’s pulse (TC-836).

Added

• One-electron calculations of atomic hydrogen from the figures: bauer_2018 (Bauer and Deng; a 2-10-2 cycle flat top at 400 and 800 nm, and against the frequency at two fixed fields, the new quantity ionization_yield_spectrum; TC-834) and le_2016 (Le et al.; sin^2 pulses at 400, 800 and 1600 nm; TC-835); Tong and Lin’s alpha and I_p for hydrogen (tong_2005). hydrogen_model, hydrogen_yields, hydrogen_spectra and cycle_pulse (sin^2 and flat-top pulses for pulse_yield) in optmat.gas.ionisation_validation; a table on the manual page (TC-833). Below gamma = 1 the static factor is 1.05 to 1.32 times every set, the candidate frequency factor 0.92 to 1.03. The curated data are now 575 source records: 7274 datasets and 102,649 values, 209 errata and 771 published formulas.

• travers_2026_trecx: argon’s one-electron ionised fraction in Wallace et al.’s 5.5 fs pulse at carrier-envelope phases 0 and pi/2 (m = 0 and m = 1), and tdse_intensity_scale_test, which puts it through Wallace et al.’s test: it matches their measured yield at an intensity scale of 1.002 (TC-836). The curated data are now 575 source records: 7278 datasets and 102,697 values, 209 errata and 771 published formulas.

• raman_response_status(gas) in optmat.gas.raman_response: the status and reason a gas’s or a mixture’s Raman response would have, without building it, or None where no constituent is Raman-active (the noble gases). It shares its rule with raman_response (solprop SEP 3, Q42; TC-840).

Changed

• The ionisation rates are three to four and a half times faster, with the same numbers to the last bit: each rate of the average over a cycle is evaluated alone at its own nodes, the nodes once, logsumexp’s arithmetic without SciPy’s dispatch, and the functions of the Keldysh parameter share their transcendentals. Argon’s default at 800 nm takes 0.2 ms a call and 3.4 us a field, from 0.8 ms and 11 us; N2’s molecular default 0.4 ms and 9 us. tools/benchmark_rates.py times them against another tree (solprop SEP 3, Q46; TC-837).

• D2’s viscosity is Assael et al.’s correlation (their hydrogen correlation with the D2 mass, times the D2 paper’s quantum factor), not a cubic spline through their Table 2. The spline wiggled at 1e-4 (the table’s rounding and its degraded 90 K cell), so no smooth series could represent it to better than 4e-4. The correlation reproduces the table within its rounding (5.1e-4; 1.2e-3 at 90 K), and a Chebyshev series of degree 20 represents it to 1e-11. The served values move by up to 5.1e-4 (1.2e-3 at 90 K); the form "table" is removed (solprop SEP 3, Q33; TC-839).

• The silica page states where fit/SiO2’s structure near 2 µm comes from: Lithosil Q2’s OH bands, which the fit keeps unscaled. They give +2.1 % of the GVD at 2 µm over Malitson’s formula, and the 2.21-2.25 µm group gives 7 % of β3. The absolute refractometry can neither confirm nor exclude them. The page also gives the model’s 1e-15 evaluation noise. No values changed (solprop SEP 3, Q37, Q32; TC-841).

Fixed

• The derivative of a solid’s index made absolute (Daimon’s CaF2 relative to nitrogen, the air-relative formulas) no longer jitters by 1e-8 between neighbouring frequencies. The reference gas’s refractivity is differenced, not its squared index. The derivatives move by up to 2.5e-8, the old error, and the indices are unchanged. FormulaModel.EVALUATION_VERSION is raised to 2, so the formula models’ digests change (solprop SEP 3, Q38; TC-838).

0.2.1 (2026-10-02)

The equations of state’s molar mass, which solprop SEP 3 needs for Gas.Air (TC-832); argon’s one-electron ionised fraction at 800 nm with a 10 fs pulse (TC-830); and a candidate Coulomb factor that depends on the frequency, as provisional models that are not the default (TC-831).

Added

• travers_2026_trecx: argon’s one-electron ionised fraction at 800 nm with the 10 fs pulse of the 400 and 1030 nm sets (m = 0 and m = 1): at gamma = 1.8 the default’s static Coulomb factor is 1.1, 1.7 and 1.9 times the calculation at 1030, 800 and 400 nm with one pulse (TC-830). One-electron runs at the all-electron calculation’s ionisation energies are archived for its comparison. The curated data are now 573 source records: 7264 datasets and 97,227 values, 209 errata and 771 published formulas.

• A candidate Coulomb factor that depends on the frequency, coulomb = "frequency": the static factor times (1 + 2 gamma/e)^(-2 n* c s), s = n* omega/2 I_p, c = 5.25 fitted to one-electron calculations of argon at 400, 800 and 1030 nm; the provisional models frequency and one_electron_frequency. Not the default (TC-831).

• EquationOfState.molar_mass and MixtureEquationOfState.molar_mass, kg/mol: the molar mass the backend’s equation uses (equal to the transport models’), and a mixture’s composition mean, for a propagation code’s mass terms. Air’s is CoolProp’s pseudo-pure 28.96546 g/mol, not the 28.9586 g/mol Lemmon et al. (2000) print (TC-832). No identity digest changes; the golden values add the molar masses.

0.2.0 (2026-10-02)

The ionisation rates validated against one-electron calculations at 400, 800 and 1030 nm, measurements of the plasma’s phase shift and Luna.jl and solprop (TC-812 to TC-829); a rate for circular polarisation; the rates’ average over a cycle by quadrature, held above the maximum field (TC-828), which changes the default rates near and above it.

Added

• travers_2026_trecx: the 6 fs one-electron yields as the ionised fraction, with convergence runs (ionised_gauss6_Ar_m0, ionised_gauss6_Ar_m1, ionised_gauss6_He; TC-812), and the static rates of Ar 3p, m = 0 and m = 1 (static_Ar_m0, static_Ar_m1; TC-813). The validator’s 6 fs yields are the ionised fraction.

• The rates of Luna.jl and solprop against the default (TC-815): a fixture of both codes’ rates and pulse yields at 800 nm, their tests on the measured yields, and a section of Strong-field ionisation rates.

• Absolute ionisation measured by plasma interferometry: nie_2022 (He, Ar, Kr, Xe and H2; TC-818) and nambu_2026 (Ar, Kr, Xe at 800 and 400 nm, linear and circular polarisation; TC-817), read from their figures, as the new quantity ionization_phase_shift; ionisation_validation.phase_shift_test, plasma_phase_shift and circular_yield compare a rate with them, and the manual states the measured yields as intensity scales (TC-819).

• The rate in a circularly polarised field, IonisationRateModel.circular_rate (with circular_enhancement and circular_saddle_point): each channel’s static rate times Barth & Smirnova’s (2013) factor for the turning field, summed over the shell’s orbitals; it fits the measured circular-polarisation yields at the intensity scales of the linear ones (TC-820).

• Static rates of Ne, Ar, Kr, Xe: bahl_2016 (one electron, from the publisher’s data files), santra_2006 (Kr, spin-orbit resolved), otobe_2004 (Xe, density functional theory); Bahl et al.’s and Santra et al.’s are static references of the validator (TC-821).

• Molecule-to-atom yield ratios and O2’s eight-photon cross-section: dewitt_2001, loriot_2008, dura_2012 (O2/Xe, N2/Ar; the new quantity ionization_yield_ratio, with each source’s detection convention) and sharma_2018 (TC-822).

• lin_2012: O2/Xe and N2/Ar at 800 and 2000 nm, the measured rise of O2/Xe with the wavelength that the default lacks (TC-823). The curated data are now 572 source records: 7248 datasets and 96,899 values, 209 errata and 771 published formulas.

• travers_2026_trecx: argon’s one-electron ionised fraction at 400 nm (10 fs, m = 0 and m = 1), a reference of the validator: the default’s static Coulomb factor is 1.4 to 3.9 times it at Keldysh parameters 1.4 to 4.2 (TC-824). The curated data are now 572 source records: 7250 datasets and 96,919 values, 209 errata and 771 published formulas.

• zhao_2016: one-electron TDSE ionisation probabilities of atomic hydrogen at 600 to 1200 nm and of He, Ne (400 nm), Ar, Kr, Xe (800 nm); the material H (atomic hydrogen) for such reference calculations. The static Coulomb factor is 1.75 to 2.2 times high for krypton and xenon at 800 nm (TC-825), and 1.09 to 1.53 for hydrogen at 1200 to 600 nm with Tong and Lin’s static correction (TC-827; TC-825 had 2 to 2.6, without it). The curated data are now 573 source records: 7260 datasets and 97,195 values, 209 errata and 771 published formulas.

• travers_2026_trecx: argon’s one-electron ionised fraction at 1030 nm (10 fs, m = 0 and m = 1): the default’s static Coulomb factor is 1.02 to 1.35 times it at Keldysh parameters 0.4 to 1.8, against 2.0 at 400 nm with the same pulse at gamma = 1.8 (TC-826). The curated data are now 573 source records: 7262 datasets and 97,210 values, 209 errata and 771 published formulas.

• The rates of Luna.jl and solprop through every test at 800 nm (Wallace et al., Nambu et al., Nie et al., Wahlstrand et al., Parker et al.’s helium), a table on the manual page beside optmat’s default (TC-829).

Changed

• The one-electron model’s agreement with the one-electron calculations is restated: 1.07 to 1.67 times their ionised fraction for argon at 6 fs, not 0.93 to 1.21 against the population beyond 20 bohr (TC-812). The default model is unchanged and, until the measured yields agree, follows Wallace et al.’s (TC-814).

• The default ionisation rate’s recorded reasons (its coulomb choice and argon’s accuracy) are restated with the new comparisons, which changes the identity digest of rate/default/*; the golden values are re-recorded, the values unchanged (TC-812).

• The accuracy reasons of the default’s Kr and Xe now cite their static references (TC-821), which changes the identity digest of rate/default/* again; the golden values are re-recorded, the values unchanged.

• zhang_2014: the excluded row of Table IX is very likely displaced cells; the Tong-Lin value is printed in the Muller column (TC-813).

• The rates’ average over a cycle is computed by quadrature instead of an expansion in the slope and curvature of the rate, which was up to 24 % low near the maximum field, made the rate turn over just above it and left rounding noise of 3e-9. Above the maximum field a statically corrected rate is now held at its value there. Each molecule’s maximum field is the largest field its alpha was fitted to or checked against (1.61 to 1.96 times the barrier-suppression field; correction_range). Golden values re-recorded (TC-816, TC-828).

• The poe test and coverage tasks run on every core with pytest-xdist: the default lane and the coverage gate in about 45 s instead of about 4 minutes; the library and fits lanes test by test. A plain pytest stays serial.

0.1.1 (2026-10-01)

OEPs 2 and 3 Final, the lazy imports, and the equation of state’s domain fixed at the melting line (TC-811), which solprop SEP 3 needs.

Added

• OEP 3 is Final (OEP 3, The molecular gases between their infrared bands): the band rules and windows, several intervals per model, the refitted molecules and the settled questions (TC-747 to TC-756, TC-770, TC-797 to TC-809).

• OEP 2 is Final (OEP 2, Compact infrared terms for the molecular gases): the compact models of the molecules and their printed forms (TC-799 to TC-801).

• Compact models of the molecules (OEP 2, Phase 1; TC-799): compact/CO, compact/CO2, compact/CH4, compact/N2O and compact/SF6, each its fit’s electronic terms and covariance with three to five effective infrared poles in place of thousands of band-rule poles, within 0.1 of the fit’s sigma in the polarizability, the GVD and the TOD over its optical intervals. Listed by registry.models, never a default. tools/fit/compact.py re-derives them after a refit.

• FittedModel.dispersion_sigma_au, the absolute one-sigma of a dispersion derivative.

• A mixture’s intervals widen across a trace constituent’s gaps (OEP 3, Q13; TC-809): where the mixture passes OEP 3’s absorption, pole and representation checks at its mole fractions, from each molecule’s new gap table (<gas>_gap_bound.csv, tools/hitran/gap_bound.py). The full CIPM-2007 air gains 1499-1516 and 1548-1562 nm.

• Infrared and static validators (OEP 3, Q19 and Q20; TC-802 to TC-807): Rollefson and Havens’s CH4 at 1.68-14.8 um, Matsumoto’s CO2 at 10.6 um, Tsankova et al.’s static CO, and Old et al.’s 1695 and 1817 nm CO2 points, compared at their fitted scale; Legay (1955, HCl) and Mitrofanov et al. (2017, air’s group delay) recorded without datasets. A validator may take a wavelength window. The fits are unchanged. The curated data are now 562 source records: 7195 datasets and 95,846 values, 209 errata and 771 published formulas.

• tools/hitran/resolution.py: the fits’ representation error inside their optical ranges at coarser resolutions (OEP 3, Q18; TC-808).

• The compact models’ printed forms (OEP 2, Phase 2; TC-801): optmat.gas.models.compact.coefficients in the polarizability form or as a Sellmeier form at a state, with the qualified intervals, the electronic covariance in the printed parameters and the Sellmeier form’s error against the full model; format_coefficients prints them to eleven digits. The tables are on The gas models, and a gallery example compares each compact model with its fit.

Changed

• Importing the gas and solid models no longer loads scipy.signal, scipy.stats, scipy.ndimage, scipy.interpolate or scipy.optimize; they load on first use, which takes optmat.gas.models.registry’s import from about 1.2 s to 0.4 s (TC-810).

• The first use of a model no longer loads the whole source catalogue (562 records, about 2 s): a lookup builds only the registry its name belongs to, for gases and solids, and the models read only the records they cite. silica_bundle() and SiO2’s index take 0.4 s instead of 2.1 s from a fresh process, the gas default_model("Ar") 1.2 s instead of 3.0 s (TC-810).

Fixed

• The equation of state’s inverse, EquationOfState.pressure, no longer fails at every density above the critical temperature where CoolProp’s melting line lies below the equation’s pressure: nitrogen to 284 K, argon to 254 K, methane to 256 K, CO2 to 328 K, air to 236 K, and helium, neon, hydrogen and deuterium at low temperatures. phase_domain now ends just below the melting pressure there. The forward law refuses states above it as solid, and the domain excludes each equation’s triple point. No value the forward law or the inverse returned before has changed (TC-811).

0.1.0 (2026-10-01)

The first version, tagged locally once OEP 1’s phases and OEP 3’s refits had merged (OEP 1, section 6.12).

Fixed

• Incomplete resolved Raman configurations now fail explicitly instead of silently becoming unresolved bands (TC-735). Invalid EUV continuum-constraint steps raise DataError before division by zero or logarithms of non-positive densities (TC-736). Existing curated Raman and EUV outputs are unchanged.

• The dispersion fit gallery now applies the fitter’s electric/refractive and infrared conventions, wavelength windows, additive corrections and uncertainty options to its direct-data comparisons (TC-729). Argon’s static dielectric point is shown at its actual residual, about 0.41 sigma, instead of the spurious 8.99 sigma; fitted error bars use the uncertainties of the fit. Stored models and fits are unchanged.

Changed

• Dry air’s mixture equation of state (TC-796; OEP 4, Q19) is the multi-fluid model at every temperature, provisional, in place of the fallback that held only at 290-320 K. It agrees with CIPM-2007’s compressibility within 8.6e-6.

• The declared scale of an uncurated like term (TC-795; OEP 4, Q17) is 2.5 cm^3/mol (was 2), the largest curated |b|. It changes only domain warnings and the mixtures’ sensitivity.

• Oxygen’s refractivity virial coefficient (TC-794; OEP 4, Q18): O2 takes Egan et al. 2026’s published estimate, b_R = 0.210 cm^3/mol at 633 nm with 50 % uncertainty, as a provisional like term (it had none). O2’s n - 1 at a state rises by 8.7e-6 at 1 atm and 9.1e-4 at 100 bar; air’s optical domain widens from 1300 to 6150 mol/m^3.

• Krypton’s refractivity virial coefficient (TC-793; OEP 4, Q18): krypton takes Achtermann et al. 1993’s optical B_R, 5.96(6) cm^6/mol^2, in place of Orcutt and Cole’s static 5.6(3). Krypton’s n - 1 at a given state rises by 1.4e-6 at 1 atm and 1.9e-4 at 100 bar; its polarizability moves by 1.2e-7 at 800 nm, within the fit’s 78 ppm.

• Neon’s refractivity virial coefficient (TC-792, TC-744; OEP 4, Q18): neon takes Achtermann et al. 1993’s optical B_R, -0.11(2) cm^6/mol^2, in place of Orcutt and Cole’s static -0.30(10). Neon’s n - 1 at a given state rises by 7.9e-6 (relative) at 1 atm and 7.5e-4 at 100 bar. Its polarizability moves by at most 1.7e-8, and the refit’s sigma at 800 nm is 8.6 ppm (was 8.7).

• The molecules refitted with their bands as rules and their windows (OEP 3, Phase 3; TC-797). CO, CO2, CH4 and N2O now:

  • carry every band as a Gaussian rule of HITRAN’s lines, and the polar molecules their rotational band;

  • hold on several qualified intervals, their optical range split only where the pure gas absorbs, plus the windows between the bands.

  CH4 now ends at 1313 nm (was 1400) and N2O at 1950 nm (was 2000). Inside the new intervals the polarizability moves by at most 0.05 (CO), 0.26 (CO2), 0.67 (N2O) and 1.9 (CH4) of the present fits’ sigma. CO2’s 9.2-10.7 um values (Marchetti) become validators, since the model is not qualified there. The static polarizabilities of CO and N2O gain their orientation part (+0.65 and +1.45 au).

• Models own their storage (OEP 1, Phase 1): their arrays are copied and read-only, and their mappings are read-only FrozenMappings, so changing an array after passing it to a model changes nothing, and writing to a model’s array raises. A gas formula’s reference density, the virial ratio and the models silica’s Raman kernel reads are resolved once per model. GasRamanResponse gains material and population_floor.

• The development gate now requires 100% statement and branch coverage in the default test lane, with the existing package scope and exclusions. Diagnostic coverage commands remain available without thresholds. Agents prefer topic branches and separate worktrees, confirming creation or switching before proceeding.

• Branch-only register numbers are explicitly provisional. poe register-sync previews and applies their renumbering and citation updates before integration, and can resolve the subsequent register append conflict without changing entries already on main.

• Removed duplicate private checks for dataset units and wavelength media already rejected by dataset construction. Public runtime guards remain and are tested in ordinary Python processes; coverage reports now include those subprocesses.

• The dedicated comparison pages and Luna comparison figure are retained as private development notes, outside the published manual and gallery. Numerical reference checks remain in the test suite.

Added

• OEP 1 is Final (OEP 1, The propagation-code interface): identity, overrides, the equation of state, transport, air, the Raman groups, the rate energies, the silica bundle, the solid tables, solprop’s baseline, the page for propagation codes and the release test (TC-745 to TC-789).

• OEP 4 is Final (OEP 4, The refractive index of gas mixtures at finite density): the refractive index of gas mixtures at finite density, its optics, density and example, with the author’s decisions on its open questions (TC-771 to TC-796).

• The release test (OEP 1, Phase 5; TC-789): poe test-release builds the wheel and sdist, installs each offline into a clean environment, and runs the quickstart, the propagation page and the public surface there. The sdist no longer picks up local tools’ state under .claude/.

• The page Using optmat from a propagation code (OEP 1, Phase 4; TC-788): what a propagation code calls, in which units, what to store and where a model holds, with snippets the tests run.

• solprop’s baseline at one commit (OEP 1, Phase 3; TC-787): every solprop fixture re-recorded at solprop d577414, with no value changed. New transport and density fixtures are compared with optmat’s on Transport of the dilute gases. solprop returns a liquid’s density for SF6 at 30 bar, where optmat refuses the state.

• A helium-argon example and the combined sensitivity (OEP 4, Phase 3; TC-786): the gallery figure of a He/Ar fill from 1 to 60 bar, and a table on Gas mixtures at finite density of what the local field, the pair terms, the density rule and the density method each change in n - 1 and the GVD.

• The density of gas mixtures (OEP 4, Phase 2; TC-778 to TC-785):

  • optmat.gas.mixture_eos.mixture_equation_of_state(composition) gives a mixture’s equation of state by one of three methods: CoolProp’s multi-fluid model, the virial fallback (each species’ own equation at its partial density plus measured cross second virials), or the partial-density rule. It provides number_density, partial_densities (the optics’ keys), pressure, compressibility, second_virial, its pairs, method, status and identity.

  • The multi-fluid models were compared with the measured cross, excess and mixture virials and the mixture densities, under a rule fixed before the comparison (optmat.gas.mixture_validation). The He-Ar, Ne-Ar and He-Ne models (Tkaczuk et al. 2020) and the air pairs’ (Gernert 2013) fail, and their mixtures take the fallback: He/Ne validated; He/Ar, Ne/Ar and dry air provisional. Dry air’s fallback agrees with CIPM-2007’s Z within 1.8e-6.

  • Seven source records: Brewer and Vaughn 1969, Martin et al. 1982, Vatter et al. 1996, Vogl and Hall 1972, Kalfoglou and Miller 1967, Blancett et al. 1970 and Kestin et al. 1981. Bose and Cole 1970’s pressure virials were added to its record.

  • A thermodynamic family of six quantities, and new pressure and molar-density units.

  • The density sections of Gas mixtures at finite density. No existing value changed.

• The optics of gas mixtures at finite density (OEP 4, Phase 1; TC-771 to TC-775): optmat.gas.mixture.mixture_optics(species) gives MixtureOptics, the Lorentz-Lorenz law of the whole mixture with its pair terms. It has the real refractivity and complex susceptibility, both derivatives, refractivity_uncertainty, every PairTerm with its status, the constituents’ common intervals and the optical domain’s density_bound. One species is the pure law exactly. No unlike-pair term is curated: each is the declared mean of its like terms. New quantity unlike_pair_refractivity_virial_coefficient. Achtermann et al. 1991’s C_R and D_R are extracted, and Achtermann et al. 1993 and Huot and Bose 1991 are curated. The new page Gas mixtures at finite density. No existing value changed.

• Several qualified intervals per fit (OEP 3, Phase 2; TC-770): fit configurations gain [model] band_nodes (the derived band rules, and the rotational system as a term class of its own, left out of electric data) and windows = true (the derived windows as the qualified intervals, with each gap’s reason in the range warning). FittedModel.representation_bound_au gives the bound on the representation error. No stored fit uses the keys yet, and every fit is unchanged.

• Refusal outside a solid table (OEP 1, Phase 2h; TC-769): TabulatedModel has out_of_range, "hold" by default as before. with_out_of_range("raise") makes a wavelength outside the knots raise StateError. The metal tables’ digests moved; no value changed.

• The silica nonlinear bundle (OEP 1, Phase 2g; TC-768): optmat.solid.bundle.silica_bundle(name), the consistent triple (n2, f_R, h_R) of fused silica for a code in the n0 = 1 convention. Bundles: default (Hollenbeck, paired), tensor (Lin and Agrawal, their f_R) and one per response. Each carries its gain check against Stolen and Ippen.

• The energy per ionisation event (OEP 1, Phase 2f; TC-767): channel_rates, energy_loss_rate, energy_per_event and energy_kind on the atomic and molecular rate models. The channel rates sum to the rate, and the energy per event is defined down to zero field. No rate changes.

• The Raman groups (OEP 1, Phase 2e; TC-766): GasRamanResponse.groups(), each damping group’s width law in SI with decay_rate(N) down to zero density, and its share of the response; line_id, a stable identifier of each line (snapshot in tests/data/raman_line_ids.json); check_densities(N); and the derivation of the two tensor parts on the manual page. The response’s vibrational band is now a short key (v1, nu1, fermi_1388, …), so the Raman digests moved; no value changed.

• Transport follow-up (TC-765): four new source records. D2’s zero-density table is explained: the D2 paper printed the universal collision functional in place of the hydrogen correlation (TC-760). SF6’s viscosity (0.3 %, 223.555-1000 K) and ideal-gas heat capacity are checked against their papers (TC-761). D2 and SF6 are now validated. O2’s heat capacity stops at 1700 K, where CoolProp’s ideal part leaves its source’s stated accuracy.

• Explicit parameters and overrides (OEP 1, Phase 2a; TC-764): model.with_parameters(**changes) on every model of the surface. It rebuilds from the class’s declared inputs, refuses derived fields, and records the changes and the curated digest in the identity, with the status user. New constructors: IonisationRateModel.from_parameters, MolecularRateModel.from_parameters, OscillatorModel.from_parameters (SI poles), GasRamanResponse.from_lines, FormulaModel.sellmeier and TabulatedModel.from_arrays. A species may be a custom name. The curated builders go through these constructors; no value or digest changes. MolecularRateModel.with_ionization_energy is now an override, so its result has the status user.

• The dilute gases’ transport properties (OEP 1, Phase 2c, second part; TC-760 to TC-763): optmat.gas.transport.transport_model(gas) for solprop’s twelve gases and CO2, with the dilute-gas viscosity, the ideal-gas heat capacity (CODATA’s R; (5/2) R for the noble gases) and the molar mass, their ranges and identity, and Wilke’s mixture_viscosity. Sixteen new source records (twenty with the follow-up) name each correlation and equation of state; CoolProp’s correlations reproduce their papers’ dilute forms to 1e-9. Ne, Kr, Xe and N2O take Huber’s dilute form (checked against the WebBook’s tables, now in data/external/nist_webbook), D2 Assael et al.’s table (provisional), SF6 CoolProp’s unchecked correlation (provisional). Two quantities, dilute_viscosity and ideal_gas_heat_capacity, in a new family, transport.

• Named air compositions and the air model (OEP 1, Phase 2d; TC-759): "dry air (N2, O2, Ar)" and "dry air (N2, O2)", CIPM-2007 reduced and renormalised; optmat.gas.air.air_model(name), the mole-fraction sum of the constituents’ recommended models with its derivative, intervals and identity, and CoolProp’s pseudo-pure dry air for its density (air_equation_of_state). The full composition agrees with Ciddor’s standard dry air (a new record, ciddor_1996) within 1.1e-8 in n from 400 to 1400 nm. The Raman response accepts the two reduced names ("dry air" is the first) and refuses the full composition.

• The equation-of-state object (OEP 1, Phase 2c, first part; TC-758): optmat.gas.eos.equation_of_state(gas), with its identity, forward law, speed of sound, phase domain and inverse, pressure(N, T), round-tripping to 1e-12; and optmat.gas.eos.pressure. The real-gas density now refuses a condensed state or one beyond the equation, raising StateError (TC-731); no curated value changes.

• The polarizability’s frequency derivative (OEP 1, Phase 2b; TC-757): dpolarizability_domega(wavelength) on every gas dispersion model, complex, in SI, analytic for every class, 0 in the static limit. Fitted models expose their reference temperature, the temperature of their infrared terms and the measured temperature coefficient (N2, O2, CO2) as metadata in their identity. O3’s first frequency derivative (dpolarizability_domega_au, dispersion_au at order 1) is now the spline’s own, which moves it by at most 4.5e-5 of its maximum.

• The molecular gases’ infrared windows, band rules and bound (OEP 3, Phase 1; TC-747 to TC-756). tools/hitran/windows.py derives, from HITRAN’s line lists, the Gaussian band rules, the folds of the weak in-window bands, the rotational systems, the qualified intervals from 12.5 um to each fit’s shortest wavelength and the representation bound, for CO, CO2, CH4 and N2O, with their provenance; loaders in optmat.gas.infrared (band_systems, infrared_windows, bound_cells, gauss_rule, fold_poles). The optical ranges are split only where the gas absorbs. The fits do not use these tables yet, and no number changes. A new source record, Golub and Welsch (1969).

• Model identity (OEP 1, Phase 1; TC-745, TC-746). Every model, response and fit of the public surface has an identity: its name, the quantity it gives, its material and sources, its status (validated, provisional, untested, comparison or user, with a reason) and a SHA-256 digest of what it evaluates, stable across processes and releases that do not touch it. Dispersion models have qualified intervals, one per model, with wavelength_range their envelope. SF6’s molecular rate is untested, and the CO2 and N2O Raman responses are provisional. The new page Identity, status and validity describes them. The public surface and golden values of every model are snapshotted in the tests. No number changes.

• Statement and branch coverage reports for the default, fit-reproduction and paper-library test lanes. New contract tests cover provenance and errata, model construction, weighted fits and uncertainty, Raman and nonlinear response, ionisation and spectral-loader boundaries.

• An About page describing AI-assisted development, paperextract, source verification and reproducible fits; clearer project motivation and acknowledgement of refractiveindex.info on the front pages.

• Links to LupoLab/optmat and an Edit on GitHub button in the manual.

• Sources for the accepted OEPs (TC-737 to TC-744), seven records of notes, no datasets: Wilke’s mixture-viscosity rule (wilke_1950, checked against the page and reproduced from its Table I), Buckingham and Pople’s dielectric virial (buckingham_1955), Garberoglio and Harvey’s ab initio refractivity virials of He, Ne and Ar (garberoglio_2020), and the mixture equations of state GERG-2008 (kunz_2012), Lemmon et al.’s air (lemmon_2000), Tkaczuk et al.’s noble-gas binaries (tkaczuk_2020) and Gernert’s humid-gas and CCS binaries (gernert_2013). An open finding: the neon refractivity virial the package uses (Orcutt and Cole’s static value) is about three times the ab initio and optical values (TC-744).

• Dispersion sources for the gas models (TC-704 to TC-724), 21 new source records:

  • helium theory: puchalski_2020 (the alpha^3 QED static polarizability, 7.5 ppb above puchalski_2016) and pachucki_2019 (the retardation term of the refractive index, a -24 ppm omega^2 term that fit/He does not yet include: open);

  • magnetic susceptibilities: lesiuk_2023b (Ne and Ar with the relativistic correction; the first magnetic_susceptibility data) and lang_2026 (N2, with its ab initio Cauchy moments at 303 K, 0.16 % below fit/N2);

  • metrology: egan_2023 (A_R of Ar and N2 at 633 nm on a piston-gauge scale, with its authors’ correction egan_2024_correction applied as errata) and the review garberoglio_2023;

  • Rayleigh scattering: thalman_2014 (N2, Ar, O2; with its 2017 erratum) and wilmouth_2019 (Ar, CO2, SF6, CH4), which resolve the methane disagreement of TC-150 (a published formula 14 % high) and show SF6 about 2 % above its fit in the UV;

  • the vacuum ultraviolet of Kr and Xe: kramer_1985 (xenon’s negative refractivity at 115-117 nm, read from a scanned figure; a check), laporte_1975, laporte_1981, gerardo_1974, and the third-harmonic papers garrett_1986, cotter_1979, junginger_1981;

  • chan_1993_erratum, applied to chan_1992 (it moves only euv/Ar);

  • method only, for compact infrared terms: voronin_2017, mathar_2004, mathar_2007, tang_1972.

• Refits with them (TC-725): Ne’s and Ar’s relativistic chi_0, N2’s first chi_0 (ab initio), egan_2023 fitted in fit/Ar (-1.3 ppm at 633 nm; Kr and Xe follow), and new Rayleigh, theory and metrology validators for He, Ar, N2 and O2. fit/CO2 moved to a lower minimum with one oscillator on its 202 nm bound (GVD -0.35 % at 633 nm, +32 % at 250 nm; its 343 nm working estimate is now 2 %): open.

• CO2’s two weak bands below 10 eV as lines (TC-726): Chan et al.’s 1Delta_u and 1Pi_g strengths as priors, the effective oscillators from 10 eV. The ultraviolet form no longer follows a modelling bound; the fit returns to within 0.3 % of its GVD before TC-725, and Bideau-Mehu’s held-out 181-254 nm values agree (chi-squared 17363 -> 57).

• A reference temperature for the molecules (TC-730): [model] reference_temperature_k and a dataset option temperature bring values measured elsewhere to it with a measured coefficient. N2, O2, CO2, N2O, H2 and D2 refer to 293.15 K; Egan’s 303 K values of CO2 and O2 are brought there (n - 1 -40 and -21 ppm).

• Helium’s retardation term (TC-728): Pachucki & Puchalski’s -3.73e-5 omega^2 a.u., added like chi_0 to the refractive polarizability (OscillatorModel.retardation_au, [model] retardation_au): fit/He’s GVD -24 ppm, and Yang et al.’s measured dispersion coefficient agrees (4.8 -> 0.6 sigma). fit/He also takes the Thomas-Reiche-Kuhn sum S(0) = 2 as a prior, which keeps its highest oscillator from running away.

• travers_2026_trecx: argon’s one-electron ionised fraction in Wahlstrand et al.’s 42 fs pulse, 17 intensities with convergence runs and the bound states subtracted, and long_pulse_yield to compare a model with it. The one_electron model is 1.24 times it on average, through its resonance structure; the one electron already exceeds the measured yield of the atom by 1.3 to 4.1 times (TC-702).

• travers_2026_trecx: the 6 fs iSurff yields are marked as lower bounds. The runs keep the Coulomb tail at the tSurff surface, which made iSurff 1.4 to 23 % low in the 42 fs runs; with the tail switched off it agrees with the ionised fraction (TC-703).

• Values read from figures (data rule 8, TC-699): where a source prints no number, its figure may be read, from the PDF’s vector drawing (the markers’ and lines’ coordinates, the axes calibrated from the drawing’s ticks) or from an embedded image, by the source’s extraction script (tools/extract/_figure.py). Such a dataset sets digitised = "vector" or "raster", names the library figure, states its digitising uncertainty, and is checked against every number the source prints; values read from an image are never fitted (the loader rejects it). pypdfium2 is a development dependency of the extraction scripts.

• vanderhart_2007 and vanderhart_2006: van der Hart’s all-electron R-matrix rates at 390 nm, the time-dependent ones of Ar (Table I, and its Fig. 4’s vector drawing) and the Floquet ones of He, Ne and Ar (read from the image of its Fig. 6, which matches the vector copy within 4 %); rmatrix_rates compares a model with them. Helium’s Floquet rate agrees with Parker et al.’s; argon’s default rate is 1.6 to 2.9 times above them near the tunnelling regime and 7 to 12 times in the multiphoton one, which is open (TC-700, TC-701).

• jagau_2018: coupled-cluster static ionisation rates of H2, CO, CH4 and O2 from the supplement of Jagau (2018), with their averages over orientations (orientation_averaged_rates(..., reference="jagau_2018")). O2’s and CH4’s correction at finite field is now fitted to them (alpha 6.1 and 18.0) instead of taking N2’s 14, and CH4’s rate takes the middle maximum of its Jahn-Teller band, 14.4 eV, where the first gave a dependence on the field no alpha reproduced. O2’s rate at 1e14 W/cm2 rises 2.8 times (TC-698).

• Enhancement proposals. Major changes to optmat are proposed, reviewed, implemented and recorded as optmat Enhancement Proposals (OEPs, docs/oeps/). There is a process document (OEP 0), a template, an archive and a consistency test (tests/test_proposals.py). Every OEP is reviewed by at least one human and one frontier AI model before it is accepted. OEPs 1 to 4 are accepted: the propagation-code interface that solprop’s SEP 3 needs, the compact infrared terms, the molecular gases between their infrared bands, and the index of gas mixtures at finite density that solprop’s SEP 11 needs.

• The curated data:

  • 557 source records: 7190 datasets and 95,742 values, 209 errata and 770 published formulas, with the provenance of every value and the evidence of every erratum;

  • the extraction scripts that reproduce them from the papers’ tables;

  • the review register.

• Gases: He, Ne, Ar, Kr, Xe, H2, D2, N2, O2, CO, CO2, CH4, SF6, N2O and O3.

  • A fitted model of the refractive index of each (dipole oscillator-strength models with HITRAN’s vibrational bands for the infrared-active molecules, and a Kramers–Kronig model from the measured absorption for ozone), with covariance and leave-one-out robustness records.

  • Real-gas densities, Lorentz–Lorenz with the refractivity virial term, and the magnetic term.

  • The complex index of the noble gases in the extreme ultraviolet.

  • The second hyperpolarizability of each gas as a process-aware dispersion model, with the conversions from the conventions of the literature.

  • Every published formula of the curated sources as a named model, and a comparison with the refractiveindex.info database.

• Materials: the records name a material from a registry (optmat.materials, data/materials.toml) in place of a gas species, so that glasses and crystals share the core. Solid datasets carry no gas state, and an anisotropic crystal’s datasets name their axis. The gases’ dependency, pyfluids, is the gas extra (TC-318).

• Solids (optmat.solid), fused silica first:

  • Franta et al.’s universal dispersion model, implemented from its published equations. It reproduces the authors’ tables to their last digit, with exact derivatives (TC-321).

  • The model interface for solids: complex index, d eps/d omega, group index, GVD.

  • Silica’s recommended complex index from 25 nm to 125 um, fit/SiO2: Franta et al.’s model refitted to the absolute refractometry of the window (Leviton & Frey; Malitson, whose indices are relative to air), which Franta’s published model misses by up to 3.4e-4 in n and 1.5 % in GVD (TC-399).

  • An index relative to room air (Edlen’s dry air at any temperature and pressure), and Leviton & Frey’s temperature-dependent formula of silica, 30-300 K (TC-398).

  • Silica’s n2(lambda) with its Raman fraction and Hollenbeck’s response (TC-393).

  • A fitting framework for solid dielectric functions (TC-322).

• The solid-state literature: 54 new source records (silica’s linear index, n2, two-photon absorption, ionisation and damage, the Raman response, compilations), with data for some sixty solids, and the refractiveindex.info solid entries and Franta’s newAD2 files as external data (TC-320, TC-325 to TC-392).

• The documentation: the manual, with accuracy pages checked against the stored fits by the tests, and a gallery of the data and the fits.

• Independent numerical references: recorded values check that optmat’s named formulas, densities, rates and Raman parameters reproduce their reference implementations (tests/test_luna_reference.py).

• The manual for publication:

  • the gallery’s figures are SVG (with PNG thumbnails), in sections;

  • the front pages and the quickstart cover the solids, the Raman response and ionisation, and a test keeps their counts of records current (tests/test_counts.py);

  • the snippets of the README and of every page that has them run in the tests;

  • a Read the Docs configuration.

• No unlicensed third-party files (TC-727): Babar & Weaver’s workbook, McPeak et al.’s data files, the NIFS tables, Franta et al.’s newAD2 files and the SCHOTT data sheet’s text are no longer in the package. They are attached to their papers in the curators’ paper library, from which the extraction scripts read them. The four newAD2 parameter sets that formulas use are held in the source records as the formulas’ terms, and the silica fit starts from franta_2016/udm_lithosil_q2.

• optmat.gas.linewidths, optmat.gas.molecular_ionisation_rates and optmat.gas.molecular_ionisation_validation load as attributes of optmat.gas, and a test keeps each subpackage’s list of modules complete.

• The ionisation rates of the noble gases (optmat.gas.ionisation_rates, optmat.gas.ionisation_validation; docs/ionisation_rates.md; TC-652 to TC-667):

  • the rate of He, Ne, Ar, Kr and Xe in a linearly polarised field, cycle averaged against the amplitude and instantaneous against the field, in SI, with the sources, range and uncertainty of each model;

  • the default model: the rate of Perelomov, Popov and Terent’ev with the static Coulomb factor, Tong and Lin’s correction of the static rate, Tong, Zhao and Lin’s coefficients, the fine structure of the ion, and an effective number of electrons set from all-electron static rates (1.47 for He, 1.0 for the others);

  • the forms of the literature as named models, for comparison (ppt_larochelle, ppt_ilkov, ppt_popruzhenko);

  • an instantaneous rate whose average over a cycle is the cycle-averaged rate (to 3 %), where the usual division by the factor of the tunnelling limit gives up to 1.4 times too much at 800 nm and 1.9 at 400 nm;

  • jagau_2016: the all-electron static rates of He, Ne and Ar from Jagau’s supplement (Tables S3 to S5), on which the effective number of electrons of Ne and Ar now rests (TC-668, TC-669); perelomov_1967 and perelomov_1968, the sources of the static Coulomb factor and of its range (TC-670);

  • wallace_2016: the 37 measured Kr+ yields of the supplement’s Table I, and the models against them (measured_points_test; TC-672);

  • optmat’s rate of molecules (molecular_rate_model(gas, wavelength), the model default; TC-677 to TC-680): the rate of the noble gases carried to H2, D2, N2, O2, CO2, N2O and CH4, with the orientation-averaged structure factors of the molecular orbital, the static Coulomb factor, Tong & Lin’s form of the correction at finite field fitted to all-electron static rates (N2, H2), the number of electrons in the ionising shell, the vertical ionisation energy, and for H2 and D2 the average over the vibrational ground state; within 3 % (H2) and 13 % (N2) of the all-electron static rates it was fitted to;

  • CO in the molecular rates, and a test of N2’s correction on it: with N2’s alpha unfitted the default is 1.26 to 1.32 times CO’s orientation-averaged all-electron static rates (thirteen angles, 0 to 180 degrees); the default is given to 1.5 times the barrier-suppression field, below which every molecule’s rate rises (it stopped rising at 1.7 with alpha 14); Guo et al.’s molecule-to-atom comparison (N2/Ar agrees, O2/Xe 4.5 times below); three figures in the manual (TC-683);

  • zhao_2011: SF6’s structure coefficients, read on the page by the maintainer’s exception (the library’s cells of the rotated table are unusable; the script checks the reading against the degenerate orbitals and the cells the library does hold), and SF6 in the default rate: 80 times N2’s at the same intensity, a prediction that nothing tests (factor 10; TC-684);

  • zhao_2011: CH4’s structure coefficients, complex and over three degenerate orbitals (combined_structure_coefficients; TC-681);

  • the molecules’ rates in the forms of the literature (optmat.gas.molecular_ionisation_rates, docs/molecular_ionisation_rates.md; TC-673, TC-674): the molecular tunnelling rate of Tong, Zhao & Lin averaged over orientations, the rate with Talebpour et al.’s effective charges, and the hydrogen-like tunnelling rate, each of one electron and as its source gives it. They reproduce solprop’s three molecular rates to rounding. None is validated and there is no default; against the orientation average of all-electron static rates (optmat.gas.molecular_ionisation_validation) the tunnelling form is 0.9 to 6.6 times the reference for N2 and twice it for H2;

  • findings: the factor (1 + gamma^2)^(-n*) in the Coulomb term of Ilkov et al.’s and Bergé et al.’s rate is derived nowhere and makes the rate too low by 2 at a Keldysh parameter of 1; Ilkov et al.’s eq. (8) is misprinted by a factor of 2; the factor exp(-2 alpha_+ F) of Luna and solprop is in no paper;

  • validation against two-electron calculations of helium (static, 780 and 390 nm), exact one-electron static rates of argon, one-electron TDSE yields of He and Ar, and the measured yields of Wallace et al. (matched within their calibration) and Wahlstrand et al. (1.7 to 8 times below the model; open);

  • new records: wallace_2016, travers_2026_trecx (the maintainer’s tRecX yields, tier 3), the yield fits of wahlstrand_2018, the fine-structure levels of the ions in nist_asd_2026; a new quantity, level_energy.

• The ionisation parameters of the gases (optmat.gas.ionisation, TC-440 to TC-444; docs/ionisation.md), for the propagation code’s rates:

  • the ionisation energies (NIST ASD for the atoms; the primary determinations for the molecules, TC-542), adiabatic and vertical;

  • the ionising orbital and its structure coefficients (Tong et al. 2002, Zhao et al. 2010);

  • n*, l* and the ADK and PPT coefficient (Ammosov et al.’s eq. (20) corrected);

  • the static polarizabilities of the neutral (the fitted models’ electronic static limit) and of the cation;

  • the model constants: Tong & Lin’s and Zhang et al.’s barrier-suppression constants, and Talebpour et al.’s effective charges.

  Validated against the ab initio static-field rates of He, H2 and N2, and Zhang et al.’s single-active-electron rates of Ar. Gas-phase cations are materials of their own.

• The Raman lines of the gases (optmat.gas.raman, optmat.gas.linewidths, TC-442, TC-492, TC-541, TC-542; docs/raman.md) of N2, O2, H2, D2, CH4, SF6, CO2 and N2O (vibrational only): every rotational S-branch and vibrational Q-branch line, with its frequency, amplitude (in solprop’s convention) and width at a stated temperature, density, composition and pump wavelength.

  • The widths are per line, against J and temperature where the data allow, with the Dicke narrowing of H2 and D2.

  • The transition polarizabilities are ab initio per J for H2 and D2 (Hunt et al.), or come from absolute cross sections. They are dispersed to the pump with Bischel & Black’s measured law, which follows Rychlewski’s ab initio dispersion to 0.8 %.

  • The steady-state gains are validated against the measured gains of H2, D2 and CH4.

• The rotational nonlinear index of the lines (TC-493): it reproduces the published rotational n2 of N2, O2 and D2 to 0-4 %. That settles the factor of two of TC-312, which lies in a paper’s printed response function.

• The Raman response functions (TC-685 to TC-696; docs/raman_response.md):

  • of the gases and dry air (optmat.gas.raman_response): h(t) and its analytic transform, with Hellwarth’s tensor form (each line’s isotropic and anisotropic share, the anisotropic part of the Q lines included), each line’s width as c + d/n + beta n with its damping group, H2’s line shift, the partners of a mixture, and the densities between which the lines are independent Lorentzians;

  • of fused silica (optmat.solid.raman_response): Hollenbeck & Cantrell’s response, normalised analytically, with the Kerr fit’s Raman fraction (the default), and Lin & Agrawal’s and Blow & Wood’s forms; the size of the kernel in a real-field model, K = 2 n0² eps0 c f_R n2, and the gain;

  • validated against measured gains, the decay of O2’s revivals and silica’s absolute gain and nuclear share (optmat.gas.raman_response_validation, optmat.solid.raman_response_validation; stolen_1973 and hellwarth_1975 curated), and compared with solprop’s responses (tools/solprop_raman_response.py);

  • the Kerr fit’s eta(tau) now uses the analytic transform of the response, and the fit takes Stolen & Ippen’s measured Raman gain as data in place of the prior on f_R (TC-696): f_R = 0.22 +- 0.03 (was 0.15, set by the prior), the long-pulse n2 2.7 % higher and its electronic part 5.4 % lower at 800 nm.

• Metals, crystals and more of the literature (TC-543 to TC-625): 67 source records.

  • Metals (a new material kind, TC-543): Babar & Weaver’s and Werner et al.’s permittivities of Ag, Au and Cu, Ordal et al.’s aluminium (51 errata).

  • Crystals: the primary indices and formulas of CaF2 (Daimon & Masumura, Malitson), MgF2 (Dodge), BaF2 and sapphire (Malitson); silica against temperature (Matsuoka, Waxler & Cleek) and in the vacuum ultraviolet (Lamy, Rife & Osantowski, Tan et al.).

  • Solids’ nonlinear data: n2 (Milam & Weber, Weber et al., Adair et al.), two-photon absorption (Liu et al.), breakdown (Du, Lenzner, Liu, Mero) and Blow & Wood’s Raman response.

  • Gases: Brion’s dipole (e,e) spectra of H2, N2, O2, CO, CO2, N2O and CH4; Meath’s DOSDs of CH4, CO, CO2 and SF6; Hohm’s polarizabilities against temperature; static dielectric virials; the CIPM-2007 air; ionisation energies, PPT and static-field rates; H2, D2 and O2 lines and widths.

• More of the metals (TC-626 to TC-648): 8 new source records and 6 extended.

  • The DESY report SR-74/7 (Hagemann, Gudat & Kunz): Al, Cu, Ag, Au and amorphous Al2O3 from the far infrared to x-rays, with 1 - n above 65-122 eV.

  • Weaver et al.’s Physics Data 18-2: Shiles et al.’s aluminium (0.04-1551 eV) and Weaver’s gold. Babar & Weaver’s analysis to 30 eV from Weaver’s website.

  • Johnson & Christy’s noble-metal films with their errors; Rakic’s aluminium; Ordal et al.’s infrared compilation (1983) and copper (1985); Leveque et al.

  • McPeak et al.’s film data files, and Stahrenberg et al.’s numbers as refractiveindex.info gives them.

• Sapphire’s two rays and N-BK7 (TC-649 to TC-651): Malitson & Dodge’s 1972 Sellmeier pair from Weber’s Handbook of Optical Materials (2003) and SCHOTT’s N-BK7 data sheet (2023, with TIE-29’s conventions), each made absolute; the data sheet’s internal transmittance (a new quantity, internal_transmittance) and dn/dT curated.

• The window crystals’ recommended models (TC-545, TC-546; docs/crystals.md): each primary Sellmeier formula made absolute with the gas the index was measured against (nitrogen, index_medium = "nitrogen", or room air). A uniaxial crystal’s recommendation is per axis.

• The metals (TC-544, TC-619, TC-646, TC-647; docs/metals.md): table models, curated datasets joined by range with a reason for each and a monotone interpolant; silver, gold and copper from 3-4 nm to 12.4 µm (copper to 56 µm), aluminium from 0.8 nm to 200 µm. A segment may be a table of 1 - n (refractive_index_decrement) or Henke et al.’s scattering factors (TC-644, TC-645).

• The composition of dry air (optmat.gas.composition, TC-548): CIPM-2007.

• The new sources in the models: Milam & Weber’s silica n2 in the Kerr fit (+0.9 %, TC-594); new validators of six gas fits, and five DOSDs’ static limits within 0.2 % of the fits (TC-606); the ionisation and Raman choices on the originals (TC-547, TC-586).

• Ionisation, Raman and solid-state sources (TC-494 to TC-540): 36 source records.

  • Primary ionisation energies of the molecules. Measured yields, which are figures only and calibrated against models.

  • Zhang et al.’s barrier-suppression constants and static rates, Talebpour et al.’s effective charges, and the Zhao et al. erratum.

  • The Weber volume (1979): Schrötter & Klöckner’s normalised Raman cross sections, and Stoicheff’s H2 and D2 and Wang’s CO2 rotational constants.

  • Hunt et al.’s transition polarizabilities of H2 and D2.

  • Bischel & Black’s dispersion of the cross sections.

  • The N2 Q widths of Rosasco et al. and Rahn & Palmer, and the D2 widths of Smyth et al. and Keijser et al. (TC-541).

  • Two-photon absorption and n2 of solids: Sheik-Bahae et al. 1991 and 1994, Bristow et al., Patankar et al.

• The ionisation and Raman literature (TC-445 to TC-491): 43 source records:

  • ionisation energies, structure coefficients, ion polarizabilities, static-field rates;

  • rotational and vibrational constants, Raman cross sections, gains and derivatives;

  • linewidths, shifts and their temperature laws.

• Raman and ionisation for solprop (TC-439): the scope and the stages of the work that lets solprop carry no material parameters. Every Raman and ionisation value solprop uses is recorded as a parity target (tools/solprop_raman_ionisation.py, tests/data/solprop_{raman,ionisation}.json), and the missing papers are listed on the roadmap.

• Further gas sources (TC-400 to TC-438): 31 source records.

  • Hohm’s 325-633 nm dispersion in the Kr, Xe, O2, N2O, CO2 and CH4 fits: its shape at 4e-5, its level free (TC-400).

  • Static dielectric polarizabilities and 633 nm refractivities as validators.

  • The absolute dc Kerr of helium and H2 (Tammer et al. 1992), CH4 and CO2 as validators of the hyperpolarizability models (TC-401).

  • Molecular photoabsorption (Brion & Thomson 1984) and dipole sums (Olney et al. 1997), for the molecules’ extreme ultraviolet.