Laboratory gas-density consistency before EOS handoff

PvtGasDensityConsistency checks whether reported gas-phase quantities can satisfy the physical identity

\[\rho_g=\frac{pM_g}{Z_gRT}.\]

Here $p$ is absolute pressure in Pa, $T$ temperature in K, $M_g$ gas molar mass in kg/mol, $Z_g$ dimensionless gas compressibility and $\rho_g$ gas mass density in kg/m³. $R=8.31446261815324$ J/(mol K). All five measurements must refer to the same gas phase at the same state and on the same molar-volume basis. Do not combine bulk two-phase density with gas Z, use liquid molar mass, or substitute a separator-gas composition for reservoir vapor. A volume-translated density also requires a Z defined from that same volume.

This opt-in arithmetic check runs without a NeqSim flash or a fit. It helps identify unit, transcription, phase-reference and source inconsistencies before calibration. It cannot establish which reported quantity is wrong. A pass means internal compatibility under the supplied bounds; it does not establish measured accuracy, independent experiment provenance, a qualified EOS or a field-history match.

Bounds and acceptance

Each immutable Quantity(value, absoluteBound) declares a closed interval. Inputs and endpoints must be positive and finite; bounds must be finite and nonnegative. Zero bounds specify point values. The caller must justify and freeze the bounds before comparison. Printed-resolution half widths are different from instrument uncertainty and from the measurement sigma used by compareHoldout.

The implied density interval is

\[\rho_{\min}=\frac{p_{\min}M_{\min}}{Z_{\max}RT_{\max}},\qquad \rho_{\max}=\frac{p_{\max}M_{\max}}{Z_{\min}RT_{\min}}.\]

Arithmetic is rounded outward after each operation. A measured-density interval passes if it overlaps this implied interval, including a shared endpoint. These are conservative box bounds; joint correlations are not modeled. Overflow or underflow that prevents a positive finite result is rejected explicitly. There is no hidden relative-error tolerance, automatic widening, extrapolation, EOS tuning or replacement of a reported value.

The evidence retains every supplied row, its whole-experiment identity, its state reference, source/page/table/row, nominal values, bounds, implied central density, implied bounds, observed-minus-implied density and compatibility. requireCompatible() throws with the incompatible row identities. Do not silently remove failing rows or relabel rows from a single depletion experiment as independent holdouts.

Executable handoff example

The following analytic tuple has an implied density of 20 kg/m³. It illustrates the API, independently of the methane EOS below. The tuple is synthetic and is not a measured methane calibration target. The corresponding JUnit regression executes these API calls and verifies both failure and wrong-sample rejection.

import neqsim.pvtsimulation.util.CalibratedPvtModel;
import neqsim.pvtsimulation.util.PvtGasDensityConsistency;
import neqsim.pvtsimulation.util.PvtGasDensityConsistency.Observation;
import neqsim.pvtsimulation.util.PvtGasDensityConsistency.Quantity;
import neqsim.thermo.system.SystemInterface;
import neqsim.thermo.system.SystemSrkEos;

Observation row = new Observation("row-1", "CCE-1", "same equilibrium gas phase",
    "synthetic analytic fixture", new Quantity(831446.261815324, 0),
    new Quantity(100, 0), new Quantity(1, 0), new Quantity(0.02, 0),
    new Quantity(20, 0));
PvtGasDensityConsistency evidence = new PvtGasDensityConsistency("synthetic sample",
    "explicit absolute bounds, not sigmas", "public synthetic data",
    new Observation[] {row});
evidence.requireCompatible();
String diagnostics = evidence.toJson();

SystemInterface original = new SystemSrkEos(300, 10);
original.addComponent("methane", 1);
original.setMixingRule("classic");
original.init(0);
CalibratedPvtModel model = new CalibratedPvtModel(original, original.clone(),
    "synthetic sample", "original v1", "tuned v1", "synthetic protocol",
    new String[] {"CCE-1"}, evidence);
String checkpointProvenance = model.getProvenance();

The overload checks the sample identity and requires all supplied rows to pass before freezing the complete JSON under gasDensityConsistency in checkpoint provenance. Serialization and held-out comparison provenance retain it. The existing constructor, or a null evidence argument, retains the previous behavior and makes no gas-density consistency claim. This gate is not automatically applied to existing models or ordinary simulation calls. Both independently calibrated EOS checkpoints remain unchanged; reservoir-volume misfit must never be used to retune them.

Continue with complete CCE/CVD/DLE/separator calibration and held-out experiments, then the pressure-path handoff and produced-phase validation. Gas-density evidence alone does not qualify those stages. Keep laboratory holdouts separate from reservoir pressure, GOR and composition holdouts.

Public primary report check

The fixture and provenance contain a small factual extract from Schlumberger DBR report 2012-10037190, Duvernay well 13-05-043-06W5, printed page 37/table 18, dated May 14, 2012. Raw psia, g/mol and g/cc values are retained before explicit SI conversion. Printed pages 3, 34 and 37 were checked visually against extracted text.

The frozen printing-resolution bounds are ±0.5 psia, ±0.5 K, ±0.0005 in Z, ±0.005 g/mol and ±0.0005 g/cc. They are not validated measurement uncertainty. An independent standard-library Python calculation uses 50-digit decimal arithmetic on the raw source units. Java results must agree within 1e-9 kg/m³; interval overlap decisions must agree exactly. No EOS parameter is used by either calculation.

Table-18 row	Implied gas density (g/cc)	Printed gas density (g/cc)	Result
Saturation, 4661 psia	0.333688435	0.339	Incompatible under printing bounds
Step 8, 1300 psia	0.075278163	0.075	Compatible
All nine rows	Eight compatible, one incompatible	Raw values retained	Entire supplied set fails the gate

The source’s executive summary additionally reports saturation Z=0.992 while its detailed CCE table reports 0.922. These conflicting source statements must be resolved before claiming an accepted laboratory benchmark. Do not replace either with a preferred value merely to improve a fit. A prior issue ledger transcribed the final CVD pressure as 1250 psia; the verified source is 1300 psia. The fixture and tests explicitly protect that correction.

The public report is a data lead and consistency case, not an output-complete laboratory-to-field validation chain. Its CVD protocol preserves gas removal and cell-volume restoration; do not replace it with unrelated constant-composition flashes. Its separator and standard bases also need explicit reconciliation before use with reservoir outputs. Passing eight rows does not authorize discarding the ninth, combining calibration/holdout rows, or inferring reservoir volumes/reserves.

Reproduce and interpret

From the repository root:

python3 src/test/resources/neqsim/pvtsimulation/gas-density/audit_reference.py
./mvnw -Dtest=PvtGasDensityConsistencyTest test

The audit prints every nominal density, interval and decision. The test compares all nine rows with the checked-in independent reference, covers analytic point and closed-interval limits, all 16 endpoint corners, nonphysical/missing/duplicate inputs, unrepresentable arithmetic, immutable serialization and checkpoint integration. Its runtime probe logs three samples after 5000 warm-up tuples for tuple construction/checking and one-row JSON export. Enable INFO for neqsim.pvtsimulation.util.PvtGasDensityConsistencyTest to view these samples; repository test logging is off by default. Wall-clock timing is evidence, not a CI threshold or an EOS speedup claim.

The 2025 Tran rescaled-CVD/density method is a separate research acceptance item. This identity check does not implement its equations, flow model or benchmark.