Predict produced composition from equilibrium reservoir phases

CompositionalReservoirHistory.Step.phaseWithdrawal adds an opt-in extraction protocol to the existing frozen-EOS tank history. Supply interval totals in mol of reservoir gas, oil and aqueous phase, in that order, the injected component amounts on the named EOS basis, and a positive explicit subdivision count. These phase totals are inputs; their component compositions are predictions. The original Step constructor still accepts measured component withdrawals. Its prescribed compositions cannot be used as independent validation predictions.

This is an isothermal, well-mixed tank with specified phase extraction rates. It does not infer relative permeability, mobility, a well productivity index, bottomhole pressure, or reservoir-phase molar rates from measured surface volumes. Those require a separate qualified flow/basis conversion model. Preserve the original and independently laboratory-tuned CalibratedPvtModel checkpoints. Changing reservoir volume never retunes the EOS.

Physical step and refinement

Each measured interval is divided into equal substeps. At substep $j$, withdraw the specified phase’s start-state equilibrium composition, add the uniform component injection, and call the existing SimpleReservoir TV-flash step:

\[\Delta n_{i,j}^{\mathrm{prod}}=\sum_{\alpha\in\{g,o,w\}}\frac{N_{\alpha,k}}{m_k}x_{i,\alpha,j}.\]

Here $N_{\alpha,k}$ is the supplied phase amount in mol for interval $k$, $m_k$ is its subdivision count, and $x_{i,\alpha,j}$ is the equilibrium phase mole fraction at the accepted substep start. The native component ledger is

\[n_{i,j+1}=n_{i,j}+\frac{N_{i,k}^{\mathrm{inj}}}{m_k}-\Delta n_{i,j}^{\mathrm{prod}}.\]

Withdrawal occurs before injection; fresh injection is not implicitly available for the same substep’s withdrawal. Missing requested phases, requests larger than available phase inventory, invalid compositions and failed TV-flash closure reject the complete replay. A zero requested amount does not require that phase to exist. No absent-phase composition is extrapolated. Restart every trial from the frozen initial state; a rejected trial does not modify the next volume trial or checkpoint.

The extraction integration is first order. Refine subdivisions until pressure, produced composition and separator outputs meet the study’s declared tolerance. The optional segregated finite aquifer is coupled on every substep; its water is separate from tracked EOS aqueous fluid and cannot be produced by selecting the aqueous phase. The model retains its fixed-temperature/immiscible-support limits.

Inspect actual production and surface products

Result.getProducedComponentMoles() returns actual interval-by-component amounts on the history’s named basis, for both input modes. It has no initial-state row. Step.getProducedMoles() is only valid for measured-component inputs; phase mode throws instead of returning fabricated zero withdrawals. Use isPhaseWithdrawal, getProducedPhaseMoles and getSubsteps to inspect the schedule.

Result.getProducedFluid(k) restores an independent EOS mixture containing the complete interval’s withdrawn amounts. Zero-production intervals have no produced mixture. Its restored tank pressure/temperature is not a measured surface state. Processing this aggregate assumes all substep/phase withdrawals are commingled before separation; separate changing substep conditions explicitly when that assumption is unsuitable.

PvtSeparatorPath freezes strictly decreasing stage pressures in bara, stage temperatures in K, and a required protocol/basis provenance reference. The last pressure/temperature pair defines the final product-volume basis. Every stage’s offgas is independently flashed to that final basis, retaining its oil, gas and aqueous products. Both liquid phases continue together to the next stage. Unsupported phases fail; all native EOS components must be accounted for.

getVolumesM3() and getComponentMoles() return final gas/oil/aqueous products in that order. getGasOilRatio() is final gas m3 divided by final oil m3; no oil gives NaN because the ratio is undefined. Never label these volumes Sm3 without stating the final conditions. No separator efficiency, entrainment or sizing is inferred. The EOS and its multiphase options remain the caller’s responsibility.

The bounded CVD/DLE generator now uses this same separator implementation, while retaining its dry two-hydrocarbon-phase requirement and fixed 1.01325 bara / 288.15 K handoff basis. This does not add wet or absent-phase depletion support.

Executable synthetic qualification

The full reproducible example is PhaseWithdrawalReservoirHistoryTest. It defines untuned PR/classic methane/n-heptane, a 100 m3 tank at 323.15 K and 80 bara, initial gas/oil volume fractions 0.7/0.3, and one 86400 s interval: 80000 mol gas and 20000 mol oil withdrawn, with 2000 mol methane and 100 mol n-heptane injected. The named array basis is deliberately reversed. The separator has 10 bara / 303.15 K followed by 1.01325 bara / 288.15 K.

Predeclared checks cover independent TP-phase extraction and direct TV-flash pressure, native component conservation, unchanged checkpoints/replay, serialization, unavailable/exhausted-phase rejection, the dilute-gas analytical pressure limit and subdivision refinement at 1/2/4/8/16. For this fixture the 8-to-16 agreement criteria are 0.5% for pressure/GOR and 0.001 absolute produced heavy-component fraction; these are example discretization checks, not field acceptance tolerances. The test logs physical predictions and elapsed runtime.

The executed example gives these conditional predictions. GOR uses the fixed final basis above; the produced heavy fraction is n-heptane mol/mol.

Substeps	Final pressure, bara	Produced heavy fraction	GOR, m3/m3
1	59.958283	0.142949	1028.8614
2	60.030818	0.146215	1000.3103
4	60.067585	0.147871	986.3514
8	60.086089	0.148704	979.4513
16	60.095371	0.149123	976.0211

A separate gas-only synthetic pressure inversion recovers 1017.2999925 m3 from a 1017.3 m3 forward case. A held-out pressure shifted by 5 bara with 0.5 bara sigma gives held-out chi-square 100.000006 and does not enter the fit. These are numerical checks with deliberately synthetic observations, not field accuracy.

PvtSeparatorPathTest also checks an independent stock-tank flash, offgas condensation, explicit final conditions, water accounting and ownership. Existing CVD/DLE generator tests guard the shared implementation. These establish analytical/numerical software behavior, not measured laboratory-to-field accuracy.

Public qualification sources and remaining acceptance

The public primary SLB Duvernay report 2012-10037190 provides CCE/CVD laboratory evidence. Printed pages 3 and 5 identify 106 °C reservoir temperature, a reported 4661 psia dew point, and standard conditions 15.56 °C / 101.33 kPa. This differs from the generalized-PVT handoff basis above. The report is a qualification candidate, not a calibrated fluid/field-history fixture or an implied redistribution license; this delivery does not reproduce its dataset or claim agreement with it.

The 2025 Tran et al. paper and its SSRN preprint remain research leads. Full-text publisher/identified SSRN delivery access returned errors during the 31 October 2026 source audit. Full equations and independent benchmark reproduction remain required; no abstract-only implementation is presented.

Issue #4740 retains public laboratory-to-field calibration/holdouts, independent measured produced-phase validation, broader depletion phase paths, multi-tank matching and research-method qualification. Pressure-volume fitting and the supplied posterior still use the existing pressure likelihood only. Compare held-out measured GOR/compositions separately on the same sample/interval/standard/separator basis; do not reuse their measurements to prescribe the produced composition or silently count them in a pressure likelihood. In-place pore volume is not recoverable reserves.