Scope and edition evidence

This audit began against NeqSim commit 22d45de95350a9b4ce88d6ba03d603743427a469 on 24 October 2026. Issue #4571 tracks completion. It distinguishes an implemented calculation type from a verified equation and from experimental accuracy over a stated domain. A class name or a positive result is not qualification.

Poling, Prausnitz and O’Connell authored the fifth edition. The publisher’s current sixth-edition record identifies Elliott, Diky, Knotts and Wilding and a 2023 publication date. It adds explicit uncertainty coverage and newer estimation/EOS material. The full sixth-edition text was not accessible during this audit. Its topic list supplies the coverage framework; this document does not assert equation-by-equation sixth-edition conformance or that the newest paper for every fluid has been identified. Mathematical correlations are independently implemented; no handbook property database or book text is redistributed.

Coverage matrix

“Present” below means source/API coverage, not full numerical qualification. Paths are relative to src/main/java/neqsim/. The nearest guides document existing choices and constraints.

Property family Existing architecture and examples Evidence and remaining work
Pure constants and characterization thermo/component/Component, thermo/characterization/TBPfractionModel and component databases Present. Establish per-parameter provenance and validity; generic molecular group-contribution estimation is not established by this inventory.
Ideal-gas Cp, Cv, H, S and formation data Component.getCp0/getCv0/getHID/getIdEntropy, SystemIdealGas Cp0/Cv0 and H/S differences now have an independent NIST Shomate matrix for methane, CO2 and nitrogen at 320/400/500/800 K from a 300 K baseline, plus differential/quadrature checks. The 2% comparison gate is limited to these states; other fluids/ranges and parameter provenance remain open. Reuses #4568 CpD coverage.
Vapor pressure, saturation and phase-change enthalpy Component.getAntoineVaporPressure and its derivative/inverse, ThermodynamicOperations saturation operations, SystemInterface.getHeatOfVaporization, Component.calculateWatsonHeatOfVaporization Pressure/derivative/inverse behavior is covered by the existing component tests summarized below; an independent NIST isopentane pressure comparison is limited to 290/298.15/301 K. Other fit ranges and parameters remain unqualified. The low-level coefficient getters now share the five-coefficient DIPPR 106 form and explicit missing-data contract; per-fluid parameter qualification remains separate from EOS phase enthalpy differences. Opt-in Watson reference scaling is checked against six benzene calorimetry points at 307-353 K only.
Pure-fluid PVT and caloric properties SRK/PR, CPA, PC-SAFT, Span-Wagner, water and hydrogen reference utilities Present. Selecting a modern EOS does not establish modern transport-model completeness. See thermodynamics.
Mixture PVT and equilibrium SystemGERG2008Eos, SystemEOSCGEos, cubic/CPA/PC-SAFT mixtures, TP/PH/PS and saturation flashes Present. Reuse #2937 for numerical flash and stability qualification. Standalone general Tsonopoulos virial prediction was not found; embedded virial terms do exist in AGA8/GERG-related code.
Activity coefficients, infinite dilution and specialized phase behavior Wilson, NRTL, UNIQUAC, UNIFAC/PSRK/UMR, solid-liquid operations, Henry-law APIs Present. Benchmark each model/parameter set and infinite-dilution limit. A supplied class does not establish a suitable parameter set for an arbitrary pair.
Density COSTALD, Rackett, Peneloux and EOS density Present; see density models. Existing correction/mixture methods should be qualified, rather than duplicated as detached utility code.
Viscosity PFCT, LBC, Chung gas, friction theory, pure-fluid reference choices Present; see viscosity models. Added: bounded pure-liquid Letsou-Stiel estimation. Repaired: CO2 background viscosity now includes the missing initial-density term and passes published Table 6/8 checks. Critical enhancement remains unimplemented.
Thermal conductivity PFCT, dilute/dense Chung, polynomial, Filippov and reference choices Repaired: dense Chung equation, units, original coefficients and conformal mixing. Filippov’s pressure factor now implements the complete CHEMPAK Missenard-derived expression with independent equation/lifecycle checks. The original tabulated-Q variant, mixture accuracy and sixth-edition conformance remain unqualified. See conductivity models.
Gas and liquid diffusivity Chapman-Enskog, Wilke-Lee, Fuller, Wilke-Chang, Siddiqi-Lucas, Tyn-Calus, Hayduk-Minhas and specialized aqueous methods Fuller known-organic atomic sums, canonical nC10 identity and invalid-input behavior are repaired, with 27 frozen independent equation values and selected CH4/He measured-fit controls at 200/250/273.15 K. Shared gas-diffusion clone storage is isolated for Fuller/Chapman-Enskog/Wilke-Lee. See diffusivity models for units, sources, ranges and exclusions. Other methods and unknown-structure fallbacks remain unqualified; dilute-gas checks establish no dense-fluid accuracy.
Surface tension Parachor, gradient theory, linear gradient theory, density-functional and empirical interfaces Present; see interfacial properties. Qualify by phase pair, critical approach and parameter source.
Uncertainty Existing numerical and study-specific uncertainty utilities General property-model validity/uncertainty contracts are incomplete. Avoid reporting a universal percentage uncertainty for a correlation or a handbook.

Ideal-gas caloric qualification matrix

IdealGasCaloricReferenceTest checks the existing Component.getCp0/getCv0/getHID/getIdEntropy methods without changing production Java or database coefficients. Frozen independent values are in src/test/resources/neqsim/thermo/component/ideal_gas_caloric_nist.csv.

The primary references are the NIST WebBook Shomate records for methane, CO2, and nitrogen, credited to Chase (1998). Nitrogen’s current record also identifies a January 2009 parameter fit. Reference calculations use the published equations and coefficients, not NeqSim’s polynomial. The Shomate reduced temperature is T/1000, Cp is J/(mol K), H is kJ/mol and S is J/(mol K). Convert reference H to J/mol before comparison. Methane uses the 298-1300 K fit, CO2 the 298-1200 K fit, and nitrogen the 100-500 K and 500-2000 K fits. At 500 K the lower nitrogen fit is used; above it the upper fit and its own F/G integration constants are used.

NeqSim’s default sensible H and temperature-only ideal entropy use a 273.15 K zero; Shomate has different integration/reference constants. Compare H(T)-H(300 K) and S(T)-S(300 K), never their unlike absolute zeros. A separate check confirms that enabling formation enthalpy changes the H offset without changing these differences. This does not requalify formation data, standard absolute entropy or real-fluid residual terms.

Fluid Target temperatures (K) Maximum Cp error (%) Maximum delta-H error (%) Maximum delta-S error (%)
Methane 320, 400, 500, 800 1.30 0.65 0.59
CO2 320, 400, 500, 800 0.08 0.09 0.09
Nitrogen 320, 400, 500, 800 0.21 0.11 0.11

The declared engineering comparison gate is 2% for each of the 36 reference comparisons. These are comparisons against an evaluated reference correlation, not new experimental measurements, universal uncertainty bands or full-range database qualification. Twelve-point differential checks verify Cp-Cv=R, dH/dT=Cp and dS/dT=Cp/T. Independent composite-Simpson integration of the public Cp getter over 300-800 K checks the analytic H/S integrals. Reference zeros, formation offsets, amount invariance and clone-state independence are tested. Existing ComponentCpDRoundTripTest, FormationEnthalpyReferenceTest, SystemFormationEnthalpyTest and SystemIdealGasTest provide adjacent coverage; these checks complement the broader worked-example effort in #3792.

No new property API, runtime route, method selection, cache, guard or calculation work is introduced. Production methods remain unchanged because this scoped audit establishes no defect requiring a fix. Other fluids, untested temperatures (including near-zero/invalid inputs), coefficient provenance and sixth-edition equation/errata conformance remain open.

Verified transport changes

Dense Chung conductivity

The previous implementation divided an already molar Cp by phase moles. For 75/25 methane/ethane at 350 K and 100 bara it returned 0.04673335678896647 W/(m K) for 1 mol and 45.777039730801135 W/(m K) for 0.001 mol. A focused regression fails against the original class and checks 0.001, 1 and 1000 mol with the correction.

The repair also restores the density contribution, correct reduced temperature, consistent SI equation, transport association factor, original conductivity coefficients and conformal mixture rules. Composition-dependent work uses the symmetric half of the binary matrix; cube roots are computed once per component. A pure-component fast path avoids mixture arrays. All evaluation state is local, so there is no composition cache to invalidate after cloning, a component edit or a temperature change. Default model selection is unchanged.

ChungDenseCorrelationTest freezes these independent Python equation evaluations with the original coefficient table. They test equation implementation, not experimental accuracy.

T (K) Molar density (mol/m³) M (kg/mol) Tc (K) Vc (cm³/mol) omega Cv0 (J/mol/K) Dipole (D) Association Conductivity (W/m/K)
300 0 0.0280134 126.2 89.4 0.0372 20.8 0 0 0.026436687902351361
300 4000 0.0280134 126.2 89.4 0.0372 20.8 0 0 0.031034910394252442
473 5810.575246949447 0.042081 364.9 184.6 0.142 82.67 0.4 0 0.059100960749808175
400 12000 0.04607 516.25 168 0.6371 90 1.69 0.175 0.11497183673220499
200 24000 0.016043 190.564 98.6 0.01142 25 0 0 0.14523150764316997

ChungDenseConductivityRegressionTest additionally checks component order, amount, repeated evaluation, cloning and physical nitrogen pressure behavior. The dense correlation is an estimator and has no critical enhancement. Mixture combining factors are unity and have not been fitted or qualified for arbitrary polar/associating mixtures.

Letsou-Stiel saturated-liquid viscosity

LetsouStielViscosity extends the existing liquid viscosity method, implements the existing viscosity interface, and is selected with PhysicalProperties.setViscosityModel. It also exposes a stateless Java API usable through the Python bridge. Units and range contracts are in the viscosity guide. It requires no new database fields or runtime dependency and is never selected automatically. Pure liquids outside 0.76–0.98 reduced temperature and mixtures are refused.

The independently published ethanol worked example is reproduced to rounding precision; tests also exercise the documented benzene phase integration, units, limits and cloning. These are implementation checks, not a qualification of compressed-liquid viscosity or of associating fluids. No pressure correction or mixture rule has been silently added.

CO₂ reference background viscosity

The baseline CO2ViscosityMethod included the dilute and residual Laesecke–Muzny terms, but omitted the linear-in-density Rainwater–Friend contribution. It failed all six finite-density rows of the paper’s Table 6. The correction restores equations 5–6 and retains the existing phase-bound model and Span–Wagner density fallback. NaN density now also triggers that fallback. Invalid scalars and unsupported species/mixtures fail explicitly.

The correlation uses the published energy scale 200.76 K, length scale 0.378421 nm, Table 3 virial coefficients, and a fixed pure-CO₂ molar mass of 0.0440095 kg/mol. The dimensionless virial coefficient is converted with Avogadro’s constant and the length scale cubed; division by molar mass is essential because the API takes kg/m³. All coefficients are fixed, with no parameter fitting in this change.

T (K) Density (kg/m³) Published background viscosity (mPa s) Baseline error (%)
220 3 0.011104 +0.13
225 1150 0.22218 +1.98
300 65 0.015563 -2.15
300 1400 0.50594 -1.42
700 100 0.033112 -3.94
700 1200 0.22980 -6.81

All six corrected values fall within half the last printed digit. The three dilute Table 6 rows, including the stated zero-density extrapolation to 10000 K, are also checked. Seven independent Table 8 recommendations additionally exercise the existing EOS, phase selection, property initialization and SI viscosity access at 240–1000 K and 1–1000 bara; the integration tolerance is 0.1%. These are checks against the published reference correlation and its recommendations, not new experimental qualification.

Critical enhancement is not included. The main static API supports 100–2000 K; the separate dilute-only API explicitly permits extrapolation to 10000 K. These scalar bounds do not establish physical phase stability or the melting limit. Callers needing near-critical viscosity must account for the missing enhancement separately.

NIST IR 8608 (February 2026), section 11 still identifies Laesecke–Muzny (2017) as the reference viscosity formulation and retains its parameters when evaluating the new alternative CO₂ EOS. That EOS is separate from this viscosity repair; it has not been implemented or qualified by this audit.

Filippov pressure-factor qualification

The Filippov liquid model (also built into the GLYCOL/AMINE/CO2WATER aqueous property sets) now uses the complete analytic expression documented in AFT’s permission-based reproduction of CHEMPAK version 4 (1998), section 8.6.5, attributed there to Danner and Daubert’s Missenard-derived procedure. The old code used 0.98+0.0079*Pr*sqrt(Tr), omitted the rational term and documented yet another equation. The conductivity guide now gives the implemented equation, units, mixture pseudo-critical approximation, unity floor, legacy cutoffs, missing-critical-data behavior and executed Java example.

FilippovConductivityMethodTest freezes ten independent Python evaluations of the published analytic expression (with the retained unity floor), over Tr=0.5-0.9 and Pr=1-200:

Tr Pr Factor Previous factor
0.5 50 1.301066653092 1.259307178569
0.7 10 1.130606599229 1.046096142096
0.8 50 1.570267830252 1.333298740445

These are equation checks, not experimental observations or an uncertainty envelope. Tests additionally cover the disable switch, state/pressure-unit changes, amount and insertion order, composition changes, physical phase/facade and specialized-set wiring, repeated evaluation and clone independence. The pure-component work array is copied on clone and resized when the model is rebound to a phase with a different component count. No cross-state cache is introduced.

The original Missenard paper and its tabulated-Q implementation are not verified by this increment. No original-paper, table-cell or sixth-edition equivalence is asserted. Mixture pseudo-critical mixing, base-conductivity coefficients, actual saturation/solid boundaries, critical behavior and physical accuracy (including at the integration-test states) remain unqualified. The 0.95 reduced-temperature cutoff is a retained discontinuity, not a physical validity bound. Nonfinite critical data produces NaN; critical data below 1e-10 retains the legacy bypass of the pressure multiplier. Neither establishes that the base polynomial is appropriate. Existing DEFAULT/PFCT paths remain unchanged; selected GLYCOL/AMINE/CO2WATER aqueous paths receive the corrected calculation. A mathematically bounded unity-floor shortcut avoids fractional powers for positive Tr below 0.95 and Pr at most 0.5; the full expression is then below 0.994.

For edition verification, the precise outstanding source is the sixth edition’s liquid thermal-conductivity pressure-correction equations, associated coefficient tables, stated applicability and any errata. A licensed extract of that section is sufficient for the next comparison; publisher scope metadata does not supply it.

Existing vapor-pressure qualification evidence

Current-master inspection on 27 October 2026 used baseline e2cb5f792f2a3480fab451bb9a161e586b0e3fda. Reuse the existing ComponentAntoineVaporPressureTest and ComponentPow10KPaVaporPressureTest; the following evidence does not require another correlation implementation. Public routes are getAntoineVaporPressure(T), getAntoineVaporPressuredT(T) and getAntoineVaporTemperature(P), with K, bar absolute, bar/K and K respectively. The component database guide documents dispatch and the retained legacy pressure scales.

Correlation/path Existing executable evidence Qualification boundary
DIPPR-101 selected by nonzero E, including legacy log/exp labels Four shipped-row pressure values; ten finite-difference derivative states; nine inverse round trips; explicit label-precedence cases Numerical equation/dispatch checks do not certify each coefficient set or fitted range.
Isopentane shipped DIPPR fit Independent NIST Antoine comparison at 290, 298.15 and 301 K, using the existing 1% comparison gate Pressure agreement at these three points only; no universal uncertainty or derivative-accuracy claim.
Explicit pow10/pow10KPa labels Synthetic pressure and analytical-derivative values; pow10KPa finite differences and inverse checks at 260/300/350 K These prescribed coefficients test scale and dispatch, not a physical substance. The pow10KPa label retains division by 100000, not a kPa-to-bar conversion.
Zero-exponent log/exp Synthetic 2-bar pressure and 1/15 bar/K derivative at 300 K Numerical implementation check only.
Wagner fallback ComponentWagnerNumericalTest adds frozen high-precision pressure/derivative and independent-pressure inverse checks, finite-difference refinement, critical/invalid neighbors and component lifecycle checks Synthetic numerical qualification at the stated points below; no real-fluid fit range or physical accuracy is established.
Domain/inverse behavior CO2 pole-extrapolation refusal; consistent pressure/derivative/inverse states at 230/250/280 K; synthetic inverse bracketing through raw values above Pc; selected missing/invalid/ionic cases These cases are not an exhaustive domain, solver-convergence or database audit.

The primary independent isopentane source is the NIST WebBook, CAS 78-78-4, Willingham et al. (1945), Antoine parameters A=3.91457, B=1020.012 and C=-40.053, with pressure in bar and temperature in K. Its stated interval is 289.44-301.74 K; all three comparison temperatures lie inside it. The record was refreshed for this audit. Agreement with that evaluated correlation is distinct from new experimental validation.

On this baseline, the two existing vapor-pressure classes pass 50 cases. The combined run with ComponentHeatOfVaporizationTest and FilippovConductivityMethodTest passes 84 cases with zero failures/errors/skips. Production equations, coefficients and APIs are unchanged by this documentation reconciliation, so it introduces no calculation cost. The inherited tests retain their tolerances; no discrepancy is fitted away.

Fit-temperature provenance, independently verified derivatives and inverses for every correlation family, systematic boundaries/lifecycle behavior, and sixth-edition conformance remain open. EOS saturation/phase enthalpy differences and the low-level latent-heat parameter discrepancy are separate contracts. This evidence complements #3792’s broader worked-example harness and does not supersede #2937’s flash campaign.

Wagner numerical API qualification

ComponentWagnerNumericalTest exercises the existing component APIs using explicitly prescribed coefficients A=-7, B=1, C=-2, D=-3, E=0, Tc=500 K, Pc=50 bar and a 100 K triple-point boundary. The isopentane component supplies an API instance only; these inputs are not an isopentane fit or a parameter set admitted for engineering use.

The mathematical function under test is the implemented form:

\[P=P_c\exp\left[\frac{T_c}{T}\left(Ax+Bx^{1.5}+Cx^3+Dx^6\right)\right],\quad x=1-T/T_c\]

Frozen independent references use Python standard-library Decimal arithmetic with 80-digit precision. Derivatives at 150, 200, 250, 350, 450, 490 and 499.5 K come from central differences with a 1e-25 K step, independently of Java’s analytical derivative. A second high-precision step verifies reference stability. At Tc, P=Pc and the one-sided derivative limit is -Pc*A/Tc=0.7 bar/K. For example, at 350 K the references are 2.90517637138374536 bar and 0.0797486275493922320 bar/K. Inverse tests consume the frozen pressures rather than Java getter output, including the critical endpoint. Pressure and derivative checks use relative tolerance 2e-13; recovered temperature tolerance is 0.001 K.

Additional cases check finite-difference refinement at 200/350/450 K, pressure-scale linearity, amounts 0.001/1/1000 mol, isolated clone coefficients, repeated/reordered state evaluation, adjacent representable critical temperatures, invalid/nonfinite inputs and an inverse target below the prescribed triple-point bracket. No production equation, coefficient, inverse setting or default changes. The batch adds no work to simulation methods and makes no timing-speedup claim.

This checks a mathematical implementation, not experimental validity, a real-fluid parameter range, near-critical physical accuracy, or original/sixth-edition conformance. Other correlation-family fit provenance and broader EOS/phase-equilibrium qualification remain separate. Reuse the existing public Java/Python component routes; no additional MCP wrapper or duplicate correlation is required.

Coefficient-based saturation latent heat

Both Component.getHeatOfVapourization(T) and Component.getPureComponentHeatOfVaporization(T) evaluate the same DIPPR 106 form:

\[H_{vap}=10^{-3}A(1-T_r)^{B+CT_r+DT_r^2+ET_r^3},\quad T_r=T/T_c\]

Temperature and critical temperature are in K, database A is in J/kmol and the result is in J/mol. The legacy getter previously treated these coefficients as a malformed polynomial in absolute temperature; the other getter omitted E. The full five-term form is documented in Aspen Physical Property Methods and Models, section “DIPPR Heat of Vaporization Equation” (vendor manual). This establishes the equation form, not sixth-edition conformance or parameter validity.

The methods return NaN for nonfinite/nonpositive temperature, T at or above Tc, nonfinite coefficients, nonpositive A, nonpositive evaluated exponent, or a nonfinite/ nonpositive result. No fit-temperature limits are stored with these coefficients; 0 < T < Tc alone does not qualify a particular fluid’s fit. The system-level EOS enthalpy-difference calculation is unchanged.

ComponentHeatOfVaporizationTest checks an independent five-coefficient value, invalid/missing parameters, temperature trends, amount invariance and clone behavior. The shipped benzene coefficients are compared at 353.3 K with the NIST WebBook value 30.72 kJ/mol (Majer and Svoboda, 1985). The shipped coefficient evaluation gives 39.147973 kJ/mol, 27.4% high: this remains an unqualified parameter set, not a passing physical-accuracy benchmark. Regression tests freeze the independent equation evaluation and do not loosen a physical tolerance to hide this discrepancy. For example, water and benzene currently share latent-heat coefficient values, so parameter provenance and fluid-specific validation remain open under #4571.

The pipeline evaporation fallback consumes J/mol directly: signed flux in mol/(m2 s) times latent heat gives W/m2. The previous extra factor of 1000 is removed, with an independent 78.295947 W/m2 equation-based benzene unit check and evaporation/condensation sign checks. The fallback still omits unavailable component correlations; that existing approximation is not a qualified substitute for coupled interphase enthalpy transport. This repair does not certify evaporation distance or two-phase thermal closure against experiments.

Watson reference-state scaling (opt-in)

A caller with trustworthy pure-fluid saturation reference data can now use Component.calculateWatsonHeatOfVaporization(T, Tref, Href, Tc) or component.getHeatOfVaporizationWatson(T, Tref, Href) through the existing component interface (also available through the Python Java bridge). The instance adapter uses that component’s Tc. The static operation is registered through runCapability. It does not use or replace the database latent-heat coefficients.

\[H(T)=H_{ref}\left(\frac{T_c-T}{T_c-T_{ref}}\right)^{0.38}\]

All temperatures are in K and both enthalpies are in J/mol. The equation and fixed exponent are from Watson (1943), equation 13, printed page 402 (original paper, DOI); the scanned equation was inspected. This reference scaling differs from NIST TDE’s fitted multi-term HVP expansion.

Require finite positive Href, 0 < Tref < Tc, and 0 < T <= Tc. The reference value is returned exactly at T=Tref and the mathematical limit is zero at T=Tc. A supercritical target, invalid reference or unrepresentable positive subcritical result throws IllegalArgumentException. Temperature differences avoid separately subtracting reduced temperatures from one near Tc. No cross-state cache is used.

The documented adapter example is executed in watsonComponentAdapterAndClone:

SystemSrkEos fluid = new SystemSrkEos(332, 1);
fluid.addComponent("benzene", 1);
ComponentInterface component = fluid.getPhase(0).getComponent(0);
double estimate = component.getHeatOfVaporizationWatson(332, 353.3, 30720);

The standalone static route with Tc=562.1 K gives 31875.128087 J/mol at 332 K. The adapter uses the shipped component Tc and therefore need not give identical digits. An independent NIST benzene comparison uses the Majer/Svoboda (1985) anchor 30720 J/mol at 353.3 K and Tc=562.1 K; none of the following Dong et al. (1988) calorimetry points were used to fit the fixed 0.38 exponent:

Temperature (K) NIST calorimetry (J/mol) Watson estimate (J/mol) Relative error (%)
307 33400 33149.24 -0.75
314 33100 32800.60 -0.90
324 32400 32291.80 -0.33
332 31900 31875.13 -0.08
344 31400 31232.93 -0.53
353 30600 30736.76 +0.45

All six points pass a declared 1% comparison gate. This is a limited benzene check, not a universal uncertainty statement or validation outside 307-353 K. Use reference data near the target temperature. The original paper identifies water below its normal boiling point as an exception. Pure-fluid phase stability/melting limits, mixtures, compressed-liquid pressure corrections and near-critical accuracy/derivatives remain outside this estimator’s qualification. The zero endpoint is a model limit, not proof of critical-region accuracy. Default EOS latent heat, all property selections and normal process evaluation costs are unchanged; work is performed only on explicit calls. The earlier database parameter discrepancy and sixth-edition evidence gap remain open.

Execution routes and computational cost

The stateless transport correlation APIs are registered in devtools/engineering_capabilities.json with parameter units, applicability, frozen examples and test sources. They use the existing runCapability static-JSON invocation route. TransportPropertyCapabilityTest executes all four routes and checks the viscosity range refusals; no new execution mechanism is introduced. The generated engineering-coverage inventory records source and evidence references, not an experimental-accuracy certification.

ChungDenseConductivityBenchmarkTest is an opt-in benchmark. On Java 17, one local run measured five batches of 50,000 calls after 20,000 warmup calls, using the same harness and an independently compiled original class from the inspected baseline. Median times were:

Components Original (ns/call) Corrected (ns/call)
1 292 157
2 290 302
8 1602 1172

These are diagnostic microbenchmarks, not stable CI thresholds or whole-process speed claims. The binary case was about 4% slower in this run, while the pure and larger-mixture cases were faster. The implementation performs the complete corrected mixture equations without an unsafe cross-state cache. Default PFCT evaluation paths are unchanged. Letsou-Stiel is an opt-in constant-cost polynomial evaluation and introduces no work in default calculations.

The CO₂ continuation was measured with CO2ViscosityBenchmarkTest: Java 17, five 200,000-call batches after 20,000 warmup calls, using the same phase-bound harness and a separately compiled original class. Median times in one local run were:

T (K) Density (kg/m³) Original (ns/call) Corrected (ns/call)
220 3 94.45 79.77
225 1150 91.48 78.18
300 65 92.13 79.30
300 1400 90.63 79.60
700 100 90.79 79.64
700 1200 90.77 81.06

The corrected model is faster in this diagnostic run despite restoring the missing physics. Residual constants are precomputed; fractional powers in the virial term use two square roots and a polynomial; integer powers use products. No mutable state cache or additional EOS solve is introduced on the ordinary finite-density path. Timing does not measure an entire flash/process calculation or establish a stable speedup guarantee.

Remaining priorities and evidence gates

  1. Obtain sixth-edition equation/errata evidence and attach per-method provenance before making a full edition-conformance statement. Do not replace a validated method merely because another method is newer; compare accuracy, range, required parameters and computation cost.
  2. Qualify CO2 critical enhancement as a separate numerical/thermodynamic increment. The background correlation is repaired and checked against the paper; the singular critical contribution remains excluded. Assess the new NIST IR 8608 alternative EOS and its potential speed/derivative benefits separately from changing transport coefficients.
  3. Qualify latent-heat parameter provenance/fit ranges and liquid pressure corrections with published examples before changing defaults. Distinguish an unsupported alternative correlation from a calculation type already supported by EOS enthalpy differences.
  4. Assess standalone Tsonopoulos and molecular group-contribution APIs; Watson reference-state scaling is now exposed with the limited qualification above. Require a genuine use case, verified parameters, finite/invalid input behavior, dimensional tests and independent values before implementation.
  5. Extend #3792’s worked-example harness across the matrix. Keep #469’s SUPERTRAPP work separate. This audit does not supersede the existing EOS, electrolyte or flash campaigns.

References

Fuller gas-diffusivity parameter qualification

The explicit Fuller route retains its existing low-pressure equation and rounded bar prefactor. The maintained organic table now uses the published additive increments rather than inconsistent literal sums: C=15.9, H=2.31, O=6.11 and -18.3 per aromatic/heterocyclic ring. Saturated cyclohexane has no such ring subtraction, and the canonical database name nC10 shares n-decane’s sum. Regressed simple-molecule volumes retain priority and are unchanged.

Accessible provenance is Gu, Cheng and Tang (2018), equation 2.3/Table 1, doi:10.1098/rsos.171936, CC BY 4.0, and the independently documented COMSOL 6.4 parameter tables linked in the diffusivity guide. These are parameter/equation evidence from later public references, not verified fifth-/sixth-edition text or errata. The frozen 70-digit Decimal values isolate the molecular lookup with prescribed masses; they do not establish physical accuracy of 27 real-fluid models.

A distinct CH4/He physical control uses the original AF/TT measured fits in Langenberg et al. (2020), Table 3, doi:10.5194/acp-20-3669-2020, CC BY 4.0, at 200/250/273.15 K and 1.01325 bara. The fixed 15% comparison gate is limited to six fit comparisons, not a universal uncertainty specification. The paper’s C2H4 rows concern ethene and are not used as ethane measurements. Known-organic arithmetic repair must not be mistaken for new experimental qualification of every affected organic pair.

The same increment isolates gas-diffusion clone work arrays and checks public fluid/phase/property evaluation, amount/composition/order, unit/scaling, invalid states, boundary neighbors and repeat behavior. No new estimator or generic specification harness duplicates #3792. Existing public component/phase Java and Python composition remains the route; no new JSON/MCP operation is needed. Unknown-structure critical-volume/mass fallbacks, component-count rebinding, other diffusion correlations, finite-concentration factors and dense-fluid validity remain separate. No constants are fitted to the control measurements.