Experimental data are available for a bounded first step in issue #4730. The repository contains 24 experimental markers extracted from a primary supporting-material figure, provenance, an offline fitting tool, and numerical checks. The resulting coefficients describe complete methane–water–promoter systems. They do not qualify component parameters for COMP or enable sH in a hydrate flash.

NeqSim’s executable hydrate models still support sI and sII. Their parameters, structure selection, and numerical paths are unchanged by this dataset.

Primary source and data basis

The source is Enlargement of Structure-H Clathrate Hydrates by New Large Molecular Guest Substances and Lattice Size Limitation, Journal of Chemical & Engineering Data (2023), DOI 10.1021/acs.jced.3c00252. The measurements appear as colored markers in Figure S3, page 4 of the supporting material.

The figure has vector markers. Their centers and axis coordinates were extracted directly: the horizontal coordinate is linear in reciprocal kelvin, and the vertical coordinate is logarithmic in reported MPa. Curve segments and legend circles were excluded. These are reconstructed observations from a published figure, rather than author-supplied numerical table entries.

The frozen files are src/test/resources/data/chemistry_benchmarks/structure_h_jin2023_digitized.csv and its adjacent .json provenance record. They retain source DOI, figure/page, source-PDF and derived-CSV SHA-256 checksums, marker coordinates, calibration, chemical names, counts, and unresolved qualification details. Decimal digits in the extracted coordinates aid reproduction; they do not establish experimental precision. The original PDF is not redistributed. Its Figshare record labels the source artifact CC BY-NC 4.0; the extracted factual coordinates are attributed.

The retrieved supplement identifies four sH systems, but does not establish a complete fluid-phase/composition basis, the absolute/gauge pressure convention, or the measurement-uncertainty budget. Pressures therefore retain the label reported MPa. No bara conversion, uncertainty weights, or inhibitor/multiphase qualification is inferred.

Reproducible preliminary fit

Run from the repository root with Python 3.11 or later:

python3 devtools/fit_structure_h_dissociation.py --output /tmp/structure-h-fit.json

This command is offline and uses the Python standard library. The output reports source_pdf_audit: "NOT_RUN" because a default run verifies the frozen dataset and its checksum, without claiming to re-read the PDF.

For a byte-identical copy of the source PDF, the optional audit re-extracts the axes and markers and compares every observation with the frozen CSV:

python3 devtools/fit_structure_h_dissociation.py --source-pdf /path/to/je3c00252_si_001.pdf --output /tmp/structure-h-fit.json

The audit requires PyMuPDF, already included in the devtools environment. The source download must match the checksum in the provenance record. No download occurs during fitting or CI.

The fitted equation is

\[\ln(P_{\mathrm{reported}}/\mathrm{MPa})=A-B/T\]

Here $T$ is temperature in K, $A$ is dimensionless, and $B$ is in K. Least squares uses log pressure and a centered reciprocal-temperature coordinate to reduce cancellation. The coefficients apply only to each complete source system and the extracted temperature interval; extrapolation is unqualified.

Large guest Points Temperature range (K) A B (K) All-point pressure AARD (%) Leave-one-out pressure AARD (%)
Iodocyclopentane (ICP) 7 275.93–286.26 35.24082 9515.04 0.933 1.448
Chlorocycloheptane (ClCHP) 6 275.97–288.98 37.12752 10097.06 1.634 2.756
Bromocycloheptane (BrCHP) 5 277.78–287.96 35.71212 9650.02 0.882 1.724
trans-1,4-Dichlorocyclohexane (t14DClCH) 6 276.28–287.48 34.25062 9264.66 1.212 1.982

AARD is average absolute relative pressure deviation against extracted markers. The largest individual leave-one-out error is 5.19%. The fixed internal holdout uses the second and penultimate temperature in each system; the remaining observations are used for fitting. Leave-one-out fitting excludes and refits for every prediction target. Both are internal checks from the same publication, not independent-study validation or experimental-accuracy claims.

The standard-library fitter reproduces a separately computed NumPy least-squares result. Tests also cover a prescribed analytical curve, narrowly spaced temperatures, invalid and rank-deficient inputs, corrupted data, coordinate calibration, a deliberately perturbed marker, and reversed pressure-temperature trends. The documented CLI runs in the test suite.

Why these coefficients cannot be inserted into COMP

COMP hydrate fields describe guest–cage interactions. Equilibrium pressure of a methane–water–promoter system is not a component/cavity Langmuir constant. The JSON always reports coefficient_type: "METHANE_WATER_PROMOTER_SYSTEM_CORRELATION" and accepted_for_COMP: false. It reports the physical component fit as blocked, even when the preliminary pressure curve has small residuals.

For a vdWP sH model, the hydrate-water reference involves all three cavity types:

\[\ln f_w^H=\ln f_w^\beta-\sum_c\nu_c\ln\left(1+\sum_j C_{jc}f_j\right)\]

Here $f_w^H$ is hydrate-water fugacity, $f_w^\beta$ is empty-host water fugacity, $f_j$ is fluid guest fugacity, $C_{jc}$ is the guest/cavity Langmuir constant, and the cavity counts per water molecule are $\nu_c=3/34,2/34,1/34$ for small, medium, and large sH cavities. Fugacity and inverse-fugacity units must match. This equation states a model requirement; an executable sH equilibrium model is not supplied by this fitting tool.

An unknown host reference, helper-gas occupancy, and large-guest occupancy can compensate for each other in a pressure-curve fit. In the high-occupancy limit, the logarithmic terms mainly constrain combinations of host and guest parameters. A low curve residual therefore cannot establish unique physical component coefficients.

The remaining physical calibration must:

  1. Fix a source-qualified sH host reference, cage geometry, and helper-gas parameterization, with consistent units and thermodynamic reference states.
  2. Establish measured compositions, phase basis, pressure reference, and uncertainty; obtain independent holdouts. Prefer numerical tables when available.
  3. Calculate water and guest fugacities using the intended NeqSim EOS/mixing rule and all relevant gas/oil/aqueous phases.
  4. Fit identifiable guest Langmuir or Kihara parameters. Check parameter sensitivity and correlation, occupancy bounds, fugacity residuals, derivatives, and independent predictions.
  5. Populate COMP only for supported components and integrate three-cavity structure selection and conserved amount calculations together. Preserve sI/sII numerical results and measured performance.

Missing data remain missing. Existing sI/sII values, copied extended-component placeholders, and the system-correlation A/B values are not promoted to sH component defaults. Issue #4730 remains open for this physical implementation and qualification.