Add Dholabhai hydrate comparisons and salt-specific accuracy discussion

[EvenSol]

The hydrate tutorial's repaired KCl solver returns a numerical boundary, but its previous external benchmark only covered NaCl. This update adds independent comparison and discussion against Dholabhai, Kalogerakis & Bishnoi (1993), doi:10.1021/je00012a045.

Comparison

• Retain all 110 non-seawater experimental endpoints across 28 pure-water, NaCl, KCl, CaCl2, NaCl-KCl and NaCl-CaCl2 recipes.
• Match Table I's measured wet-basis recipes and Tables III-IV's absolute pressures and temperatures; retain original units and source-table provenance.
• Add every-point tables, family and per-recipe bias/MAE/RMSE, two comparison figures and three CSV exports.
• Explain the pressure-search endpoint, instrument specifications versus combined uncertainty, paper regressions versus model predictions, salt-specific discrepancies, freezing/precipitation limits and the separate Burgass benchmark.
• Match the 10.02 wt% KCl endpoint at 11.30 bara and compare 10.00 wt% and finite-CO2 controls.

All 110 points converge, conserve input and meet the saturated GAS+AQUEOUS comparison requirements. This is numerical acceptance, not an experimental accuracy claim. Matched K10 predicts -2.546515 °C versus measured -4.130 °C, a +1.583485 K error. NaCl RMSE is 0.324862 K, KCl 2.685082 K, and concentrated KCl errors reach 6.729004 K. Per-recipe metrics keep opposing mixed-salt biases visible.

The actionable accuracy limitation is tracked by equinor/neqsim#4234, under #3144. No parameters are fitted and no empirical mud correction is applied to these measurements. The five synthetic-seawater endpoints are excluded because their complete salt recipe is outside the three-salt builder. The supplied ACS PDF and original figures are not redistributed.

Validation

• Final clean Python/IPython process and fresh JVM: 24/24 code cells, 22/22 numerical/workflow checks, 197/197 accepted hydrate attempts, 253.4 s; outputs and all nine figures retained.
• Source-pinned NeqSim master snapshot f59810115c07d9a3ef479e55be45a5972963231d, source-built Java 3.23.0, Python bridge 3.20.0. Verified reused JAR SHA-256 e26d9dd82e86d351fa58fe05e04801e2af6aebb17d19081c690736d924536f19; cached source files matched the clean immutable checkout, and class origin was asserted.
• Current core master 17c0beca5a5ca53401fa214c2a542892e5a6a467 was inspected; no thermo/hydrate/electrolyte/database changes from the tested pin. That newer head was not rebuilt.
• Focused scripts/check_notebook.py --require-main-source with the updated ledger and affected-link catalog: 0 errors, 0 warnings.
• Repository checker unit tests: 6 passed. All 330 catalog notebook links resolve against the connector-fetched repository tree.
• AST/schema, code readability (<=100 characters), sequential execution counts, complete retained outputs and Colab math checks pass.
• nbconvert HTML / MathJax 3.2.2 SVG / WeasyPrint / Poppler review: every Markdown and result cell, 8 display + 24 inline expressions, 9 figures, all tables and 31 notebook + 13 catalog rendered pages inspected.
• No pre-commit configuration exists on the inspected Colab base; its PR workflow runs the notebook checker and unit tests.
• Hosted Colab installation/build and repository-wide CI were not executed locally. VALIDATION PENDING CI.

Documentation impact

The existing notebook gains the measured comparison, scientific discussion, sources, exports and accuracy follow-up. Its catalog description and maintenance ledger are updated in the same branch. No NeqSim Java API, model equations, parameters or defaults change.

Open this draft in Colab