The Henry columns in COMP.csv are backed by 83 selected rows in src/main/resources/data/HenryWaterSource.json. Each row records the component, CAS identity, source solubility constant, temperature slope and reference number. The matching bibliography is HenryWaterReferences.bib in the same directory. HenryWaterCoverage.csv inventories all 389 rows: 83 imported correlations (79 distinct database identities plus four exact-identity aliases), 35 qualified reference-temperature-only points, 22 remaining literature candidates, 140 estimated/other-source candidates, 59 ionic rows, 49 without an exact CAS match in the archive, and water itself. A candidate match is a research lead, not validated data. Rows without a dispatched correlation contain zero coefficients and represent unavailable correlation data. Zero coefficients are not a physical constant, and the former 900,0,0,0 overflow sentinels have been removed.

Source and limitations

Source: Rolf Sander, Compilation of Henry’s law constants (version 5.0.0) for water as solvent, Atmospheric Chemistry and Physics 23, 10901–12440 (2023), doi:10.5194/acp-23-10901-2023. The compilation is distributed under CC BY 4.0; this is a selected and converted subset, not the original database. Source values are from Hsbp.f90 in the version 5.0.0 download.

Selections use literature-review entries (type L) with a reported temperature slope. The selected reference is explicit; the compilation can contain other, conflicting values. These values have limited precision. The current selection covers neutral hydrocarbons, common gases, refrigerants and selected solvents. Neutral data are never assigned to a charged component merely because it shares a CAS number. Reactive acids, amines, hydrated species and reference-only data without slopes need separate qualification. No claim of complete coverage is made.

Four legacy PVTsim-named rows (methanolPVTsim, propanePVTsim, ethanolPVTsim and nbutanePVTsim) have the same CAS number, molecular formula and InChIKey as their qualified canonical component. They therefore reuse the same molecular Henry expression and source record. The inherited type-L record is Sander reference 3500, Burkholder et al. (2019), JPL Publication 19-5. The implemented numerical subset remains attributed to the CC BY 4.0 Sander compilation; no JPL report text is reproduced. The machine-readable record gives no numerical uncertainty or primary experimental range, so published precision and the local 298.15 K van’t Hoff scope are retained. This does not assert that their other pure-component parameters are identical. The two MEG PVTsim rows remain unavailable because the canonical MEG row has not yet passed source, definition and range qualification.

The neutral n-pentane and i-pentane rows are exact CAS and InChIKey matches to Sander’s pentane and 2-methylbutane records; NeqSim’s nC5 and iC5 formula labels are abbreviations for C5H12. The selected type-L rows are the highest-ranked slope-bearing Brockbank (2013) value for n-pentane and the only slope-bearing type-L value, from Plyasunov and Shock (2000), for isopentane. The raw values are respectively 7.3e-4 and 7.9e-4 mol kg^-1 atm^-1, with local slopes 3900 and 3000 K at 298.15 K. Neither Sander row contains a numerical uncertainty. Other type-L rows provide an explicit source spread: n-pentane values span 7.3e-4 to 8.8e-4 with slopes from 3400 to 3900 K; the alternate isopentane point is 7.4e-4 and has no slope. These comparisons are not fitted uncertainty intervals.

The Brockbank thesis is publicly readable but its record identifies only an institutional copyright policy, not a permissive reuse license; the Plyasunov article is publisher-copyrighted. NeqSim reproduces only the numerical facts from the CC BY 4.0 Sander compilation. Both expressions remain local van’t Hoff descriptions about 298.15 K (and 0.1 MPa where stated), not qualified finite extrapolation ranges or independent new regressions.

Elemental mercury is an exact CAS, formula, neutral-charge and InChIKey match to Sander’s water-solvent record. The selected type-L JPL Evaluation 19 row is Hsbp = 0.13 mol kg^-1 atm^-1 with a 2600 K local slope at 298.15 K. The first independent measured row, Andersson et al. (2008), reports the same value and slope at compilation precision; Sanemasa (1975) reports 0.13 and 2500 K. This agreement is validation, not a refit or uncertainty interval. The Sander machine row reports no numerical uncertainty or finite experimental range, so the implemented expression remains a local van’t Hoff approximation. Only numerical facts from the CC BY 4.0 compilation are reproduced; no JPL or publisher text is copied.

The two-parameter expressions are local van’t Hoff approximations about 298.15 K, not newly fitted experimental data. Their individual experimental temperature ranges and uncertainty are not qualified by this import. Tests at 288.15, 298.15 and 308.15 K verify arithmetic and derivatives, not experimental accuracy throughout that interval. Do not use this coverage as an unrestricted high-temperature, brine, mixed-solvent or reactive-absorption model. Existing qualified IAPWS selection remains preferred for supported aqueous gases.

Qualified reference-temperature-only points

HenryWaterReferencePoints.json contains 28 exact-CAS neutral hydrocarbon points from the type-L entries attributed by Sander to Plyasunov and Shock (2000), doi:10.1016/S0016-7037(99)00330-0. That source evaluates hydration thermodynamics at 298.15 K and 0.1 MPa. The machine-readable Sander rows do not encode a numerical uncertainty, so the catalog retains their published precision and states the uncertainty limitation rather than inventing one. The set covers selected branched C6-C9 alkanes, C10-C14 n-alkanes, cycloalkanes, pentenes/heptenes and alkylbenzenes. It excludes reactive species, ions, aliases and estimated (Q/E) rows.

Six additional type-L points attributed by Sander to Brockbank (2013) cover 4-methylheptane, cis-2-pentene, cis-2-heptene, heptylbenzene, octylbenzene and nonylbenzene. The selected raw Hsbp values are respectively 2.7e-4, 4.5e-3, 2.4e-3, 2.7e-2, 1.9e-2 and 1.5e-2 mol kg^-1 atm^-1. Brockbank’s public thesis describes critically evaluated recommended values and group-contribution methods at 298.15 K and 100 kPa; these six values are retained only as type-L reference points, not characterized as new measurements or NeqSim fits. The ScholarsArchive record links institutional copyright terms rather than a permissive data license, so NeqSim reproduces numerical facts only from the CC BY 4.0 Sander compilation.

The exact-identity 4-ethyltoluene row adds one type-L point from Mackay and Shiu (1981), Sander reference 479. The Sander machine row reports Hsbp = 0.20 mol kg^-1 atm^-1 at 298.15 K, equivalent to Hm = 5.06625 bar kg mol^-1. Independent vapor-pressure/aqueous-solubility rows span 0.16 to 0.20 mol kg^-1 atm^-1 at compilation precision. This spread is validation evidence, not a fitted uncertainty interval. Neither the machine row nor the compilation entry reports a total reference pressure; the catalog therefore stores that field as unavailable instead of assigning 0.1 MPa. The point remains exact-temperature-only and does not define a slope or derivative. Mackay and Shiu’s review is publisher-copyrighted, so only numerical facts from the CC BY 4.0 Sander compilation are reproduced.

Each catalog row can override original-reference citation, URL, rights, uncertainty, identity basis, and point conditions. Exact CAS and molecular identity are required. Four rows also match the Sander InChIKey exactly; the two cis-alkene NeqSim rows have the same connectivity block and exact cis CAS/name but omit the stereochemical InChIKey layer. That limitation is explicit rather than silently treated as an exact key match. None of the six rows receives a temperature slope.

HenryWaterReferencePointCatalog exposes immutable lookups by exact CAS number or the exact NeqSim component name. A point stores source identity, convention, units, reference temperature and pressure, bibliography, license, uncertainty note and validity statement. Its temperature-taking getter succeeds only at exactly 298.15 K; all other temperatures and every temperature derivative return NaN. When the source does not report a total reference pressure, the pressure getter also returns NaN; this does not alter the reported partial-pressure Henry convention. These points are intentionally not copied into COMP.csv, do not make hasHenryCorrelation() true, and do not enter GE, Pitzer or IAPWS dispatch. A single value therefore cannot silently become a constant polynomial with a fake zero slope.

Fail-closed candidate dispositions

HenryWaterCandidateDispositions.json records the component-specific decision for all 22 exact-CAS literature candidates that are not admitted as correlations or reference points. The file records water as solvent, source record and type inventory, identity assessment, Henry-definition risk, original-source rights boundary and the evidence required before admission. It contains no coefficient, reference value, fitted slope, pressure assumption or uncertainty estimate.

The dispositions are deliberately more specific than a generic candidate label:

The Sander v5.0.0 compilation is the CC BY 4.0 identity and source-inventory basis. Underlying publications retain their own rights and must be reviewed for species definition, units, pressure, temperature range and uncertainty before numerical facts are adopted. Every disposition remains fail-closed: COMP.csv, HenryWaterSource.json, HenryWaterReferencePoints.json and all model dispatch are unchanged. The audit checks that all 22 rows remain outside both correlation and reference-point catalogs and that the two MEG aliases provide no independent evidence.

Fail-closed estimate and other-source dispositions

HenryWaterEstimateDispositions.json records all 140 exact-CAS rows whose Sander inventory status is estimate_or_other_source_requires_review. These are research leads, not approved numerical data. The manifest carries each component name, CAS, formula, available InChIKey, source-record URL, source-type codes and a scientific admission boundary. It contains no Henry value, fitted slope, assumed pressure, uncertainty or validity range.

The component-specific inventory is partitioned into four fail-closed groups:

Five duplicate-identity aliases are explicit: glycerol/TEG, H2SO4/sulfuric acid, HNO3/nitric acid, and the mojibake aliases for NH2OH and N2H4. Aliases provide no independent data. The audit requires all 140 rows to match COMP.csv and the coverage inventory exactly, remain neutral, stay outside both numerical catalogs, and carry no numerical Henry fields.

Sander v5.0.0 supplies the CC BY 4.0 identity and source-type inventory. Underlying publications retain their own rights; source-type codes do not establish permission, intrinsic/effective convention, units, pressure basis, uncertainty or range. A single later-qualified value must use the existing reference-temperature-only semantics rather than an invented zero slope. COMP.csv, IAPWS paths, GE/Pitzer dispatch, and calibrated reactive models are unchanged by these dispositions.

Fail-closed unmatched-identity dispositions

HenryWaterUnmatchedDispositions.json records all 49 rows for which the reviewed Sander v5.0.0 source archive has no exact-CAS record. No source publication or numerical value is selected for these rows. The manifest carries the database identity, component role, exact archive-search result and the evidence boundary that would have to be crossed before any future admission.

The inventory separates materially different reasons for staying fail-closed:

A stored zero ionic charge does not turn a salt, reaction-model aggregate, condensed phase or pseudocomponent into an independently volatile neutral molecular solute. Future admission requires exact identity, water as solvent, an intrinsic Henry definition, traceable source rights, units, conditions, uncertainty and a finite validity range. A qualified single value must use the existing reference-temperature-only semantics; a temperature correlation requires suitable independent data, residuals and validation.

The manifest contains no coefficient, fitted slope, reference value, pressure assumption or validity range. The audit requires all 49 rows to remain outside the correlation, reference-point and prior disposition catalogs, to match COMP.csv and the coverage inventory exactly, and to preserve their explicit component-role boundary. COMP.csv, IAPWS paths, GE/Pitzer dispatch and calibrated reactive models are unchanged.

Units and equations

The source reports $H_s^{bp}=m/p$ in mol/(kg atm), with slope $B$ in K:

\[H_s^{bp}(T)=H_s^{bp}(298.15)\exp[B(1/T-1/298.15)].\]

The historical component polynomial includes a water-molar-mass factor. Its actual reference is on the molality scale, despite older Javadocs saying bar:

\[H_m(T)=1.802\exp[h_0+h_1/T+h_2\ln T+h_3T].\]

Here $H_m=p/m$ has units bar kg/mol. Conversion from the source is exact apart from its published rounding:

\[H_m(T)=1.01325/H_s^{bp}(T).\]

Thus $h_0=\ln[1.01325/(1.802H_s^{bp}(298.15))]+B/298.15$, $h_1=-B$, and $h_2=h_3=0$. At infinite dilution in water, the mole-fraction reference is $H_x=H_m/M_w$, using $M_w=0.01801528$ kg/mol. Generic GE fallback references perform this conversion; Pitzer retains $H_m$ and its existing molality activity factor. Desmukh-Mather and Kent-Eisenberg retain their empirical calibrated reference convention: changing that convention independently of reaction constants breaks reactive VLE benchmarks. Their standard-state requalification is separate campaign work. The derivative getter returns $dH_m/dT$, not $d\ln H_m/dT$.

API and compatibility

Integrated EOS gas/oil and GE liquid references

The hybrid architecture assigns the gas and hydrocarbon liquid to EOS phases and the aqueous or polar liquid to its selected GE phase. Henry data describe a solute in a specified solvent; they are not a pure-fluid equation of state. For a water-free generic GE liquid with an applicable vapor-pressure correlation, the pure-liquid reference is used even if the database tags the component as a solute. Supercritical gas and dense hydrocarbon/CO2 cases use their model-owned EOS roles. A Pitzer liquid remains a water-solvent model: a pure methanol liquid requires a suitable liquid model rather than a water Henry coefficient.

Dissolved hydrocarbons in water use a Henry solute reference even when their pure-fluid database row is tagged as a solvent. Molecular formulas retain this decision when phase initialization has replaced the hydrocarbon classification. Missing Henry data retain the explicit finite unsupported-solute limit; the fallback does not establish hydrocarbon solubility.

For a generic aqueous GE solute, the implemented reference is

\[\phi_i=(\gamma_i/\gamma_i^{\infty,w})H_{x,i}/P.\]

The pure-water infinite-dilution calculation retains the owning phase’s component indices and interaction parameters, including user-set NRTL parameters. Other species are reduced to numerical traces in an isolated clone. It does not reload a different binary model from the database or mutate the source composition. The corresponding temperature derivative includes the reference normalization:

\[\frac{\partial\ln\phi_i}{\partial T}=\frac{\partial\ln\gamma_i}{\partial T}-\frac{\partial\ln\gamma_i^{\infty,w}}{\partial T}+\frac{\partial\ln H_{x,i}}{\partial T}.\]

NRTL publishes its analytical activity derivative. Models without an analytical activity derivative use isolated fixed-composition differences for that activity term. Pitzer neutral species use their molality activity directly, so this symmetric-to-Henry normalization is not subtracted from their derivative. Desmukh-Mather and Kent-Eisenberg differentiate their existing empirical Henry references. Desmukh-Mather includes its water Poynting pressure and temperature terms; both amine models’ constant ionic fugacity coefficients have zero pressure and temperature derivatives. These corrections do not change reaction constants, parameter qualification, or Henry data admission.

GeneralEosGeFluidTest exercises pure-fluid and multicomponent phase roles; SystemHybridEosGeFlashTest covers phase restoration, reordering, reactive conservation and process composition. GeHenryReferenceTopologyTest and ReactiveGeDerivativeConsistencyTest compare dilute-reference and fugacity derivatives with independent analytical limits and fixed-composition differences. Numerical consistency does not replace experimental qualification of a GE parameter set, a mixed solvent, a reactive Henry definition, or pressure effects absent from the selected model.

Data corrections change raw coefficient values and unsupported-solute behavior. The generic GE fallback conversion also corrects the former standard-state mismatch. Applications that relied on placeholders or assumed that this raw API returned a mole-fraction constant need migration. Pitzer brine/reaction fits require regression review when their underlying Henry reference changes.

Reproduction and remaining work

Run python3 devtools/check_henry_water_data.py to check every compiled row, identity, exact-alias equivalence, provenance coverage, point-only separation and polynomial conversion. Supply the downloaded henry_5.0.0_f90.zip as an argument to additionally verify each selected correlation and reference point against the original Fortran source. HenryWaterDatabaseTest checks database loading, source values, finite-difference derivatives, standard states and absence. HenryWaterReferencePointCatalogTest checks exact-CAS identity, immutability, point-only evaluation, unit conversion, cloning and serialization.

Issue #4044 tracks the remaining component-by-component source review, reference-only data, experimental range and uncertainty qualification, reactive-species distinctions and application-level validation. New regression fits must include underlying data, units, residuals, range, provenance and independent holdout evidence; do not fit missing slopes from a single reference value or hide an unsupported species behind a placeholder.