SystemElectrolyteCPAstatoil uses a potassium-specific CO2 short-range interaction to reproduce the sign and magnitude of KCl salting-out near the single calibration state. The change corrects the previous strong salting-in response. It is not an empirical hydrate-temperature offset and does not alter the hydrate model, phase-equilibrium tolerances, or experimental fixtures.

Parameter and model semantics

The existing electrolyte-CPA Born-diameter convention is retained. An attempted global change to the Solbraa (2002) Appendix C characteristic diameters was rejected because exact-base controls showed that it changed NaCl reactive-state conservation and methane/NaCl high-salinity hydrate behavior, systems outside this KCl-specific scope. The Onda calibration below therefore applies to the current NeqSim model convention and must not be transferred to a different Born formulation without refitting and independent validation.

The symmetric CO2/K+ coefficients are W0 = -2.2981e-4 m3/mol, W1 = 0 m3 K/mol, and W2 = 0 m3/mol. The mixing rule evaluates

\[W_{ij}(T)=W_0+W_1(1/T-1/298.15) +W_2[(298.15-T)/T+\ln(T/298.15)].\]

These are short-range electrolyte-CPA coefficients, not dimensionless cubic-EOS binary interaction parameters. The CO2/Cl− pair retains its legacy W0 = 1.27e-4 m3/mol. KCl data constrain the sum of the two gas–ion contributions; retaining chloride is therefore a fixed allocation convention, not an independent ion measurement.

The specialized default applies only to PhaseElectrolyteCPAstatoil, and only while both shared CO2 gas–ion constants retain their library defaults. Explicit pair setters and explicit changes to either shared constant take precedence. ScRK and other electrolyte-CPA phase implementations retain their existing defaults.

Calibration source

The only calibration target is the Onda et al. (1970) Table 3 correlation at 298.15 K:

\[k_s=-0.2277-0.0362+0.3416=0.0777\ \mathrm{L/mol}\]

in the source convention $\log_{10}(\alpha_{water}/\alpha_{salt})=k_s I$, with solution molarity $I=c_{KCl}$. The executable check evaluates infinite-dilution CO2 at 1 mol KCl per kilogram of water and converts to solution concentration using the modeled physical density. Onda’s value is a global least-squares ion contribution derived from several gases and salts, including compiled literature data; it is not a direct raw CO2–KCl series.

W1 = W2 = 0 is a zero-order temperature extrapolation. Onda alone does not qualify that extrapolation over the 269–363 K holdout range.

Independent holdouts

No He–Morse (1993), Yasunishi–Yoshida (1979), Dholabhai et al. (1993), mixed-salt, or Burgass point is used to choose the coefficient. Repository diagnostics keep those sources separate:

Passing numerical checks establishes state construction, conservation, and the correct salting-out direction. It does not show agreement within experimental uncertainty. The generated CSV diagnostics report residuals without converting them into a relaxed acceptance tolerance.

On the implementation head, the six 273.15 K He–Morse states decrease monotonically with KCl and have 29.38% absolute relative deviation on the reported molality basis. All seven Yasunishi–Yoshida concentrations are matched; their Ostwald coefficients decrease monotonically and have 17.69% absolute relative deviation. The eight Dholabhai hydrate holdouts give K10 RMSE 0.179 K (maximum absolute error 0.298 K) and K15 RMSE 0.247 K (maximum absolute error 0.348 K). These figures describe this exact model head; they are not acceptance tolerances and remain subject to independent review.

The salt-only KCl water-activity diagnostic remains distinct from the dissolved CO2 correction. At 273.15 K it differs from the Archer reference by −0.00215 for K10 and −0.00467 for K15 before CO2 is added. The saturated Burgass NaCl control retains five points within 1 K and a sixth error of −1.121 K at 194.77 bara.

Electrolyte-CPA association iterations reject non-finite Hessian corrections before changing site fractions, allowing the volume solver to retry without poisoning the final brine state. This numerical safeguard does not refit the interaction parameters or relax the component-balance and fugacity checks.

Full-family Dholabhai benchmark

CO2BrineHydrateFullFamilyAssessmentTest retains 110 measured endpoints from 28 recipe labels, not just the eight K10/K15 states. The fixture co2_brine_hydrate_dholabhai1993_full.csv preserves Table I wet-basis mass percentages, original temperatures in K, pressures in MPa and Table III/IV membership (54/56 points). Family counts are pure water 4, NaCl 25, KCl 21, CaCl2 21, NaCl-KCl 17 and NaCl-CaCl2 22. Recipe labels such as Na10-1/Na10-2 retain their distinct measured compositions; nominal concentrations are not substituted. The fast test freezes the complete fixture hash and checks all eight overlaps against the existing K10/K15 fixture.

Source: Dholabhai, Kalogerakis and Bishnoi (1993), DOI 10.1021/je00012a045, Tables I, III and IV, pp. 651-652. All adopted numerical values were visually checked against the user-supplied paper and reconciled with the prior executed Colab transcription. The ACS article is copyrighted: only factual values and source identifiers are retained, not the PDF, figures or article prose. The fixture is a numerical benchmark transcription, not a claim that the article has an open license. Pressure is converted by 1 MPa = 10 bar. The paper added atmospheric pressure to corrected gauge readings, so no second atmospheric correction is made. Reported RTD accuracy of +/-0.06 K and gauge accuracy of +/-15 kPa are instrument specifications, not a combined endpoint uncertainty or an accuracy pass band.

The unchanged nonreactive electrolyte-CPA/GF construction uses 1 kg water, 10 mol CO2, mixing rule 10 and the actual three-salt recipe, with independent cold construction and a 278.15 K hydrate-search guess at every pressure. Salt amounts use the total water-plus-all-salts denominator. The output retains calculated minus measured temperature, aqueous water-activity proxy, dissolved CO2 fraction, guest and water fugacities, phase labels, native saturation, original-feed closure and numerical disposition for every endpoint. Per-family and per-recipe bias, MAE, RMSE and maximum error are recomputable from accepted raw rows; rejected rows remain present and invalidate the final numerical assertion. Opposing mixed-salt recipe errors must not be hidden by a pooled signed mean.

Numerical acceptance reuses the existing hydrate endpoint gates: GAS+AQUEOUS, hydrate residual <=1e-6, molecular log-fugacity residual <=1e-8, reconstructed component and original-input mole-fraction residuals <=1e-10, phase charge <=1e-10, composition normalization <=1e-9, phase-fraction normalization <=1e-12 and nonaqueous ion mole fraction <=1e-40. Completed and failed rows are written before the final assertion. There is no experimental-error assertion and no fitted parameter, temperature correction or new tolerance. The previous 28-state contribution assessment and Burgass saturated/finite-inventory controls remain separate and unchanged. Model-only water-activity/fugacity splits are not independent high-pressure observations or causal temperature corrections.

Five synthetic-seawater points are excluded explicitly because Table II requires additional species beyond this three-salt construction; they are not replaced by NaCl. Ice stability, salt precipitation, hydrate kinetics, MEG and commercial-mud qualification remain outside scope. Extending coverage to 110 points does not itself qualify the model or close independent scientific review.

The initial assessment on production base 015fdaa24914c0c8cb587846e6da7a4594fc56fc passed all 110 numerical endpoints, with the following experimental residuals (calculated minus measured):

Family Points Bias / K RMSE / K Maximum absolute error / K
Pure water 4 0.363967 0.383012 0.524395
NaCl 25 0.158030 0.324867 0.777350
KCl 21 0.097114 0.259109 0.539193
CaCl2 21 0.444887 0.760814 2.145498
NaCl-KCl 17 -0.907341 1.278908 2.286861
NaCl-CaCl2 22 -0.015232 0.510562 1.186522

The concentrated Na7K10 and Na15K5 recipe biases are -1.619736 K and -2.013034 K, respectively; KCl-only agreement does not qualify mixed brines. Ca20 has +1.611812 K bias and a +2.145498 K maximum error. The near-zero NaCl-CaCl2 family bias hides opposing Na5Ca15 (+0.559971 K) and Na15Ca5 (-0.944962 K) recipe biases. These observed discrepancies remain validation evidence, not new fitting targets or acceptance tolerances.

Salt-free CO2 baseline discriminator

CO2WaterBaselineReferenceAssessmentTest compares nineteen evaluated liquid CO2 fractions from the IUPAC–NIST CO2/water evaluation, system 62_1, Table 1 Model column, at 273.15–363.15 K and 101.325 kPa CO2 partial pressure. This includes the 323.15, 348.15 and 363.15 K temperatures of the He–Morse KCl series, so salt-free absolute bias can be separated from its KCl salting slope without fitting a production parameter. These are evaluated model values, not raw measurements. The fixture co2_water_carroll1991_evaluation.csv records the source, units and reuse basis; only nineteen factual values are retained, not an article or figure. Carroll, Slupsky and Mather (1991), DOI 10.1063/1.555900, primary Table 4 was checked at overlapping temperatures. The displayed web Eq. (1) has a normalization inconsistency with its table and is not used.

The nonreactive electrolyte-CPA calculation uses 1 kg water, 10 mol CO2, explicit zero K+/Cl-, mixing rule 10 and multiphase checking. Total pressure is solved with a bracketed secant/bisection iteration until y_CO2 * P = 1.01325 bara within 1e-10 bara. Each trial flashes a fresh fluid with the same construction and feed inventory; this avoids reusing a collapsed phase state during pressure changes. Trial states must contain gas and aqueous phases. Total pressure and guest fugacity are reported separately. Three constructions per temperature retain baseline, explicit database reset and reversed component insertion order: 57 calculated records from nineteen independent reference states, not 57 independent observations. apparent_f_over_x_bara is a finite-composition endpoint ratio, not a newly qualified infinite-dilution Henry constant.

Frozen numerical gates are molecular log-fugacity residual <= 1e-8, original component balance <= 1e-12 (relative for molecular components; absolute mol for zero-ion inputs), phase-fraction/composition normalization <= 1e-12, charge residual <= 1e-12 mol, partial-pressure mismatch <= 1e-7 bara and identical-repeat CO2 mole-fraction difference <= 1e-10. Returned phases must be GAS+AQUEOUS. All completed records are written before the final numerical assertion, including when a later trial raises an exception; an incomplete assessment remains failed. The test supplies no physical-accuracy pass band; signed relative reference residuals are retained even when numerical gates pass.

Hosted run 40226079125 retains all 57 rows in electrolyte-assessment-evidence-1. Its manifest records PR head 57ccb4131352cd32b3fd4f44620bfdcdf46d663a, merge checkout be6c8130e01ec69a3252e7d97d67db61c35d45a5, one executed producer test, no failure/error/skip, complete selected evidence, and qualification: not_assessed. All rows are GAS+AQUEOUS and pass the frozen numerical gates. Maximum partial-pressure mismatch is 6.26e-11 bara; molecular log-fugacity residual 1.85e-9; component-balance residual 3.56e-16; repeat mole-fraction difference 1.90e-19; and construction spread in calculated CO2 mole fraction 4.23e-18.

The calculated salt-free CO2 mole fraction is below the evaluated reference at all nineteen temperatures. Signed residuals range from -27.30% at 273.15 K to -14.59% at 343.15 K; MAPE and signed mean magnitude are 17.99%. The residual improves through the mid-temperature region, then reaches -16.15% at 363.15 K. At the matched 323.15, 348.15 and 363.15 K He-Morse temperatures, salt-free residuals are -15.13%, -15.01% and -16.15%; the corresponding mean residuals across each KCl block are -16.51%, -16.39% and -18.07%. Thus the dominant absolute deficit is already present in the salt-free CO2/water baseline. KCl still controls the separately retained salting slope and high-molality curvature, including local salting-in; hydrate compensation remains an application-level possibility.

This comparison is diagnostic rather than a direct residual subtraction: Carroll reports liquid mole fraction at 1.01325 bara CO2 partial pressure, whereas He-Morse reports molality at its source partial pressure. It therefore does not uniquely identify a neutral binary or association parameter. The salt-free guest fugacity is 1.00985, 1.01353 and 1.01787 bara at the three hot matched temperatures, respectively; its modest variation does not by itself separate gas-fugacity from aqueous-model contributions.

Runtime diagnostics identify rule ClassicSRKT2, CO2 2A/two sites and water 4C/four sites, cross beta 0.075 and cross energy 8327.5 J/mol. Stored zero cross energy selects the combining rule; it does not disable association. The default binary coefficients give k(T) = -0.27686 + 0.001121 * T[K]. These are observed legacy values, not a source-authenticated recommendation. Complete joint parameter provenance and independent scientific review remain required before changing this model.

The explicit-zero-ion construction is deliberate. The separate 21-record ion-omitted/ordinary-CPA precision diagnostic retains ten component-balance failures at the same 1e-12 gate. This passing subset neither erases those failures nor qualifies other constructions. Full-family experimental qualification and independent review of #4367 remain open despite the expanded diagnostic coverage; no legacy parameter, hydrate equation or tolerance is changed here.

Matched-temperature water and guest fugacity partition

CO2BrineHydrateMeasuredStateAssessmentTest reuses the same source-audited 110-point fixture without adding measurements or changing any recipe. At each original measured temperature and absolute pressure it independently calculates the actual brine and a salt-free CO2/water comparator, both with 1 kg water and 10 mol CO2. The nonreactive electrolyte-CPA, mixing-rule-10 and native GF hydrate conventions are unchanged. The comparator contains dissolved CO2; it is not a CO2-free pure-liquid reference.

Let faq and fhyd denote aqueous and hydrate water fugacity in bara. The dimensionless quantities are:

At common T/P and hydrate structure the GF empty-lattice and pressure prefactor is shared, so G isolates the model’s guest-fugacity/occupancy contrast. Structure indices are retained rather than silently assuming a common structure. A is a model aqueous-water fugacity contrast, not an independently measured salt-only water activity. R_water still combines neutral CO2/water and hydrate-reference limitations; this algebra does not uniquely identify a physically incorrect parameter or validate the shared hydrate reference. No residual is converted into an empirical temperature offset.

The assessment also independently cold-repeats every brine and calculates its own native hydrate root from a 278.15 K search guess. It then cold-checks the fixed-state calculation at that root. This gives 440 fixed-state fluid checks plus 110 native root searches; no CSV from another producer is an input. The native setFug() reference initialization is followed by unit hydrate-water composition, matching the production temperature operation. A regression test deliberately substitutes the feed water fraction and verifies that the same root-consistency gate rejects that finite but invalid reference.

The CSV retains all 110 pairs, the original measured T/P, independently calculated root temperature, raw water and CO2 guest fugacities, dissolved CO2 mole fractions, selected structures, phase identity, original-feed/total and reconstructed closure, charge, molecular fugacity, normalization, ion leakage, phase trial distances, root residual, repeat difference and disposition. CO2 mole fraction is not molality. Known measured inputs and any raw root temperature remain in full-width failed records. All outcomes are written before the final numerical assertion; an exception or numerical rejection cannot become accepted physical evidence.

Frozen gates remain GAS+AQUEOUS, finite non-negative normalized fluid states, original-feed/total/reconstructed fractional closure <= 1e-10, charge <= 1e-10, molecular log-fugacity <= 1e-8, beta normalization <= 1e-12, composition normalization <= 1e-9 and nonaqueous ion fraction <= 1e-40. Identity closure and the independent cold repeat are <= 1e-12; native-root 1 - fhyd/faq is <= 1e-6. Hydrate equality is intentionally not required at measured temperature: that residual is the diagnostic observable. No experimental-error pass band or small-G assertion is introduced.

The prior executed investigation on master 9d6366f3af86ac79522a533b6c7da3696fc4b310 found structure I for all paired states, maximum absolute G about 6.65e-6 and maximum abs(G/A) about 7.68e-5 across 106 saline points. These are observed model-only contrasts, not physical qualification or frozen fitting targets. Na7K10/Na15K5 residuals remain negative and Ca20 positive. Independent mixed-brine activity/solubility evidence, complete joint neutral-parameter provenance and independent #4367 review remain necessary before a physical correction. Existing seawater, ice/precipitation, kinetics and commercial-fluid exclusions still apply.

CO2-water donor-incidence diagnostic

The Dholabhai assessment reports the molecular association probabilities needed by the physical-permittivity model without changing the production model. The definition follows Coelho, Franco and Firoozabadi (2025), equation 22: a molecule-pair quantity is a selected-site bond incidence. No implementation or fitted parameter is copied from that work.

NeqSim’s existing topology assigns water the 4C site signs {+,+,-,-} and CO2 the acceptor-only 2A signs {-,-}. Let X_D be the mean of the two water donor-site fractions and X_A the mean of the two water acceptor-site fractions. Because a water acceptor can bond only a water donor in this water/CO2 topology, its mass-action balance gives

P_ww = 1 - X_A
P_wc = X_A - X_D
P_w  = P_ww + P_wc = 1 - X_D

Here P_ww is water-water donor incidence, P_wc is water-CO2 donor incidence, and P_w is total water-donor incidence. The assessment records both site fractions, all three probabilities, and the within-pair donor and acceptor symmetry residuals for every K10/K15, pure-water, NaCl, CaCl2 and mixed-salt state. The exact algebraic identity is checked at 1e-14; this is an identity check, not an experimental-accuracy tolerance.

These columns isolate the molecular-association input that the future Maribo-Mogensen physical-permittivity state will consume. They do not calculate that permittivity, alter Debye-Huckel or Born, introduce a CO2 dipole, change a parameter, or qualify hydrate predictions.

Exact-head K10/K15 result and causal boundary

Hosted run 40114797409 executed the producer at PR head 1041835c41c8b9a5e2e664b9d7b65f75673e44ad (merge checkout 6ec42cc872630a5a94ce8522c071f11cf4d809b6). The 28-row CSV is in electrolyte-assessment-evidence-0; shard 3 did not select this producer. Its manifest records one executed test, zero failures/errors/skips, complete selected evidence, and the exact source/output hashes.

The four requested Dholabhai endpoints are:

Recipe Pressure / bara Measured / K Calculated / K Error / K Model suppression from matched salt-free state / K
K10 11.30 269.02 269.317673 +0.297673 3.543717
K10 34.85 277.86 277.775116 -0.084884 3.720364
K15 14.15 269.03 268.958338 -0.071662 5.762735
K15 35.75 276.01 275.661797 -0.348203 5.998445

Across those pairs, adding KCl lowers the water-activity proxy by 0.04843–0.07757 and dissolved CO2 mole fraction by 0.00155–0.00467, while total water-donor incidence increases by only 0.00106–0.00188. The incidence split changes through a larger water-water increase partly offset by a lower water-CO2 incidence. CO2 guest fugacity changes by only 0.032–0.540 bara. These are matched-state model contrasts, not measured contribution decompositions. Association incidence is therefore not the dominant scalar change in this diagnostic, although a nonlinear permittivity model could still amplify it.

The present sub-kelvin K10/K15 agreement is not caused by this diagnostic PR. It first appears after merged PR #4367, which introduced the CO2/K+ W0 default. Independent inspection of Onda et al. confirms the source’s base-10, molarity-based additive correlation and the tabulated components -0.2277 - 0.0362 + 0.3416 = 0.0777 L/mol at 25 °C. It also confirms that those values are a global least-squares gas/cation/anion decomposition, with an overall reported standard deviation of 0.052, rather than a unique direct measurement of a CO2/K+ short-range coefficient.

Accordingly, the executable 298.15 K target is source-consistent, but the mapping to W0 = -2.2981e-4 m3/mol is model calibration. Holding the legacy CO2/Cl− value fixed is an allocation convention, not an independently identifiable ion-pair result. W1 = W2 = 0 remains an unsupported temperature extrapolation. PR #4367 received only COMMENTED reviews (one automated style review and one author response), no accountable scientific approval, before merge. Its strong Dholabhai KCl holdout is useful evidence but does not close the independent provenance gate, especially with the retained He–Morse/Yasunishi deviations and mixed-brine family/recipe biases.

He-Morse pressure-driver and solvent-basis repair

Run 40168078608 failed at HM363-KCL3000: the old pressure driver returned 0.342881516 bara against a 0.304988250 bara CO2 target. Independent trials showed that its low pressure bracket included salt-depleted gas-only states and some intermediate two-phase states failed molecular fugacity equality. The 10 mol CO2 / 1 kg feed-water construction also changed aqueous salt molality through evaporation.

The revised diagnostic brackets from conserved GAS+AQUEOUS trials and matches both gas partial pressure and final aqueous solvent mass. It uses a 1 mol CO2 reservoir and compensates feed water to retain 1 kg aqueous water. An independent half-reservoir regression at the failing hot concentrated source point checks that this choice does not control the resulting solubility or pressure. Every trial verifies component inventory, aqueous-ion closure and molecular fugacity equality. The CSV records actual KCl molality and numerical residuals, preserving completed rows if a later case fails. See the benchmark notes for the exact numerical gates. Historical runs using fixed feed water are not like-for-like evaluations of the final aqueous source molality.

Local validation on NeqSim 3.26.0 with OpenJDK 17 executed all 26 source rows and the independent reservoir regression (2 tests, no failures/errors/skips). The maximum partial-pressure error was 9.91e-11 bar, the maximum aqueous KCl molality error 4.70e-10 mol/kg water, component-inventory residual 7.11e-15 mol, and molecular log fugacity-ratio residual 3.51e-12. The formerly failing HM363-KCL3000 state converged at total pressure 0.935796555 bara and CO2 partial pressure 0.304988250 bara.

These numerical checks do not establish experimental accuracy: calculated CO2 solubility remains 11.4-31.5% below the source values, and the 4.998/5.000 molal endpoints at 323.15, 348.15 and 363.15 K still show local salting-in. Those model deviations remain explicit evidence for the independent scientific review.

This corrects the assessment boundary conditions and failure reporting. It does not change production flash equations, interaction parameters, source values or experimental-accuracy criteria, and it does not establish model qualification.

Retained numerical evidence

The Core Build, Test, and Javadoc workflow retains an electrolyte-assessment-evidence-<shard> artifact from every Java 21 slow-test shard, including after failed test runs. Download the artifacts from the same workflow run; the eight assessments may be distributed over different shards. Surefire reports remain in their existing, separate artifacts.

Each bundle includes a manifest.json and only the allowlisted assessment CSVs produced by tests selected for that shard:

Assessment CSV data rows (excluding header)
Dholabhai K10/K15 plus pure-water and cross-salt controls 28
Archer salt-only, pressure-only and CO2-addition sequence 10
He–Morse dissolved CO2, complete five-temperature series 26
Yasunishi–Yoshida Ostwald coefficients 7
Burgass saturated and finite-inventory NaCl controls 12
Carroll evaluated salt-free CO2, three setup constructions 57
Full Dholabhai family, all supported measured recipes 110
Matched measured-state aqueous/guest partition, same Dholabhai recipes 110

The combined bundle has 360 diagnostic rows, not 360 independent measurements. The two 110-row assessments deliberately use the same 110 measured endpoints.

The manifest records the actual checkout commit separately from the GitHub event SHA and (when present) PR head SHA, run/attempt/shard context, selected test classes, producer Surefire counts and JVM properties, and SHA-256 hashes of outputs, producer source, reference fixture and this provenance page. Archer reference values are inline in its producer test, rather than in a separate fixture. Retrieve the referenced sources at the recorded checkout commit to inspect the source/access basis.

CSV bytes, headers, units, residuals and intentional NaN fields for non-measured control states are retained unchanged. Row counts describe the existing diagnostic matrix; they introduce no accuracy criterion. A selected producer that passed but did not generate its CSV fails evidence packaging. Malformed or truncated output is retained with an explicit packaging error. Missing, skipped, failed and not-yet-completed producer tests cannot establish complete evidence. Outputs from unselected tests and unrelated files under target/ are not collected.

selected_evidence_complete applies only to the selected assessments in that shard, not the full eight-assessment matrix or the rest of CI. A shard with no selected assessment has a count of zero. Check every producer across all shards and the original test results before using a combined matrix. Every manifest keeps qualification as not_assessed: artifact retention and structural validation do not supply independent thermodynamic review, a causal contribution split, agreement within experimental uncertainty, or engineering approval. Artifacts remain subject to the repository’s GitHub Actions retention policy.

Qualification boundary

The correction also affects CO2 with potassium and counterions other than chloride; those systems are not qualified by the KCl calibration. Temperature dependence, high ionic strength, mixed solvents, mixed salts, and hydrate-model accuracy require independent evidence. Pure-water CO2 bias is not fitted or compensated here.

Primary sources