TwoFluidPipe Evidence Matrix and Supported Envelope

This page is the qualification boundary for TwoFluidPipe. It separates available code from numerical verification and public experimental qualification. A completed calculation, finite profile, small pressure change, or stationary pressure trace is not by itself evidence that the model converged or represents the requested physics.

Evidence labels

Label Meaning
Implemented The public configuration and runtime path exist. This says nothing about convergence or accuracy.
Numerically verified An executable test checks relevant conservation, convergence, boundedness, sensitivity, or refinement contracts.
Experimentally qualified A reproducible public measured-data comparison passes its declared error gates, with source provenance and uncertainty or acceptance definitions.
Not qualified A required numerical or experimental gate fails, is unavailable, or has not been run. The result must not be promoted by loosening the gate.
Unsupported The configuration is rejected because the governing boundary or state information is missing.

Current evidence matrix

The transaction/execution update at 1f65f683 passed 422 focused tests in 62 classes, including four separately selected slow coupled/reference tests. Fifteen riser and one coarse-gas five-second gates failed separately at that revision. The later outlet-consistency repair clears the coarse-gas gate and promotes it to ordinary CI; the fifteen unsplit riser gates remain unsuccessful, with a countercurrent annular/slug force discontinuity isolated in a failing replay. The subsequent coupled-predictor repair restores all five coupled-pressure progress regressions without rejected substeps. These counts distinguish the verified runtime contracts below from the incomplete flow qualification.

Capability and configuration Implemented Numerically verified Experimentally qualified Evidence and use boundary
Positive-flow steady gas/liquid pressure, holdup, thermal and terrain profiles Yes Yes No general claim Require the complete steady convergence report to be converged, every residual below its recorded tolerance (including total phase mass flux below getMassFluxTolerance()), and no pressure-floor or wall-clock termination. Repeat mesh sensitivity for the actual geometry.
Steady gas/oil/water on the compact 3 km, 10-degree uphill fixture Yes Yes No The 30/60-cell results differ by 0.538% in arrival pressure and 0.983% in mean liquid holdup. All three phases remain present and the final thermodynamic/holdup reconciliation is inside the unchanged 1e-4 tolerance. The unavailable historical 73.8 km case is not covered.
Liquid-rich unchanged-boundary transient with shared slug force balance, interfacial pressure, and coupled pressure/momentum Opt-in Yes Not applicable Over 1,800 s, inventory drift is 1.323% at 40 cells and 1.358% at 80 cells, below the declared 2% fixture gate, with total-mass closure. This does not qualify slug loads or another operating envelope.
Default liquid-rich unchanged-boundary transient Yes No No The recorded 1,800 s inventory drift is 5.757%, above the unchanged 5% gate. Do not infer default-mode qualification from the opt-in shared-force result.
Mohmmed et al. public horizontal air/water slug kinematics Harness and data implemented Conservation only Failed At 40 cells and 0.05 s outer steps, 3/9 comparisons pass, MARE is 1.0402, and maximum absolute relative error is 3.2147. The 40/80-cell and 0.05/0.025 s sweep is non-monotone; steady starts are unconverged and pressure-floor limited and transients clamp outlet backflow. The fixed 20% speed and 30% length/frequency gates remain unchanged.
Conservative severe-slugging characterization against the public Tengesdal envelope Opt-in Five-second coupled progress restored Failed The earlier 600 s run gave 65.163 kPa, below the 68.6 kPa lower gate, without required repeated settled cycles. Centered pressure traction restores the five-second shared-closure and subcell-force regressions while retaining the valid bubble-domain guard. Short numerical progress does not qualify the experimental amplitude, period or sustained-cycle gates.
Flash-driven phase appearance/disappearance with phase and energy ledgers Yes Yes No Check gas/oil/water and total mass, transfer closure, temperature sensitivity, and latent-inclusive energy balance. Record EOS, mixing rule, composition, pressure, temperature, relaxation time, mesh, and time step.
Named-component advection for supported positive-flow boundaries Yes Yes No Require every named-component ledger, bounded normalized phase fractions, phase/component synchronization, and component-sum closure.
Positive trace-phase component inventory at synchronization Yes Yes No TwoFluidComponentTraceInventoryTest checks repeated closed substeps at 1e-11 to 2e-10 kg, conservative condensation through the 1e-10 kg synchronization allowance, and round-off handling without creating components. Larger phase/component mismatches reject without ledger mutation. The synchronization allowance must not erase transported components.
Coupled closed-face momentum response Yes Short-time analytical limit No TwoFluidClosedBoundaryMomentumTest checks Euler and RK4: a finite pressure impulse over 1e-8 s retains adjacent physical-cell inertia, while both integrated external mass fluxes are exactly zero and total mass closes. Legacy uncoupled boundary handling is unchanged.
Component-resolved downstream outlet and rejected component-substep isolation Yes Yes No The published component flows equal accepted outlet transfers divided by interval duration, including after TP reflashing. Unequal gas/oil/water transport, delayed composition fronts, reordered names, zero flow and failure isolation are tested. General unsplit component advection remains open.
Complete legacy transient transaction Opt-in Bounded regression coverage No Failed intervals preserve accepted hydro/component/thermal/slug state, reports, histories and clocks. Connected streams, upstream storage, thermal calculators and layers retain identity. Other owned submodels are new snapshots after acceptance. Concrete SRK/PR/SRK-CPA/SRK-CPAs phases are supported; unaudited EOSs and pipe subclasses reject.
Frozen-phase unsplit execution through runTransient Opt-in Bounded regression coverage No Always stages complete intervals and verifies accepted phase ledgers. Original SRK/PR phase compositions and density branches persist across calls. Spatially nonuniform/changing phase composition, reverse phase outlet flow, energy/phase transfer and tracked slugs remain unsupported. Preparation-only diagnostics have a broader signed-flow scope.
Five-second gas execution and uniform three-phase fixed point Opt-in Selected configurations No Four-cell gas runs at 0.1 s and eight/sixteen-cell runs at 0.05 s pass consecutive accepted calls with unchanged nonlinear/interval tolerances of 1e-10/1e-8. Consistent outlet phase fractions repair the single-phase Jacobian kink; the coarse case now runs in ordinary CI. These are continuation/conservation checks, not spatial-accuracy qualification. A separate closed uniform three-phase fixed point passes.
Inclined annular film eligibility, transition blending and trace-phase Jacobian Opt-in criteria; derivative correction enabled Local branch and analytic derivatives No The captured 0.594-liquid-holdup conflict converges with the film constraint. A separate opt-in annular/slug force blend removes the gas-lift point switch and advances the historical backward-Euler 16-cell gates from 0.663/0.688 s to 1.451/1.671 s; the 24-cell gate reaches 0.95 s. All still fail before five seconds. Phase-relative probes retain trace drag/volume derivatives. Neither local criterion implements full film/droplet/reversal physics.
Current unsplit Tengesdal five-second matrix with film constraint Harness implemented Failed No After the outlet repair, corrected-face 16/0.1, 16/0.05 and 24/0.05 cases stop near 1.303, 1.881 and 1.736 s. All six historical/face cases reject their prefixes and are unchanged by the inclined transition blend, showing that their blockers occur earlier. The transition-blended backward-Euler historical lane advances materially but all three cases still fail line search. Later closure/domain transitions, trial velocity guards and unsplit 180/600 s qualification remain open.
Explicit finite-volume face terrain Opt-in Yes No N+1 elevations, actual cell arc lengths and midpoint pressure with external-face offsets. Signed gravity/energy and constant-density hydrostatics are checked under nonuniform refinement. This is not general transient well-balancing.
Conservative slug/film + named components + phase transfer + thermal balance Yes Yes No A closed four-cell wet-gas cooling case over 0.05 s closes all ledgers on 0.05/0.025/0.0125 s outer partitions. Adjacent-grid water/heat sensitivity is below 4%; marker-displacement sensitivity decreases from 3.14% to 1.99%. Marker length/age are partition-invariant, and transactions reproduce the same-grid result. See the model guide for numerical values. This is coupled-ledger evidence, not spontaneous slug initiation.
Reverse outlet inflow with named-component transport and no external composition Unsupported Fail-closed No Configuration is rejected in either setter order. The last interior composition is not a physical external boundary condition.
Multi-stage conservative slug + component + phase-transfer coupling Unsupported Fail-closed No Phase appearance inside an intermediate stage needs stage-local component inventories. The coupled four-way path is currently restricted to single-stage Euler.

The Mohmmed source provenance, exact spreadsheet cells, point selection, missing-data policy, and error definitions are maintained in TwoFluidPipe Reporting and Validation. Those fixed engineering gates are not claimed as measurement confidence intervals; the paper does not publish pointwise uncertainty intervals for the selected observations.

Supported-use envelope

Use TwoFluidPipe only when the case fits a row above and its runtime evidence is retained:

  1. Run steady initialization and inspect the complete immutable convergence report. A stationary pressure profile is insufficient if liquid split, holdup, thermodynamics, or total pressure drop remains outside tolerance, or if total phase mass flux does not close against the inlet.
  2. For transient work, require requested elapsed time, phase and total mass balances, positivity, and all sticky pressure/momentum, pressure-limiter, rejected-substep, outlet-backflow, component, and energy diagnostics applicable to the selected configuration.
  3. Repeat mesh and time-step sensitivity at the actual length, diameter, inclination, composition, phase split, thermal boundary, rates, and event duration. Evidence from one fixture does not define a universal operating envelope.
  4. Treat seeded slug markers as transport/coupling probes. They do not demonstrate spontaneous slug initiation, sustained cycles, arrival statistics, or load distributions.
  5. Use PipeBeggsAndBrills for a correlation-based steady or quasi-steady screen, not for conservative distributed line-pack dynamics. Use a separately qualified transient model or a controlled experimental study when the requested TwoFluidPipe row is failed or unsupported.

High-throughput steady mass conservation (#3686)

TwoFluidPipeSteadyMassFluxTest reproduces the 3,100 m well with a 0.23 m diameter, 2,380 m rise, 20 sections, 205 bara and 90 degrees Celsius inlet, and the issue’s synthetic SRK fluid and stock-tank rates. The original 166.113217788 kg/s case lost 3.473846596 kg/s at the outlet section because gas velocity was capped at 100 m/s. At 5% higher flow the deficit was 7.007383459 kg/s; the 5% lower-flow case did not reach the cap.

The regression checks all three rates against phase-summed section mass flux with relative tolerance 1e-10 and absolute tolerance 1e-12 kg/s. It also requires the original hydraulic and thermodynamic residuals to pass. Separate gas, oil and water fixtures exercise the former steady gas/liquid caps, and injected 2% and nonfinite reporting errors must prevent convergence even when the hydraulic residuals have settled. The existing 30/60-cell three-phase fixture remains the free-water and mesh-refinement check. This is conservation and numerical-convergence evidence, not experimental or critical-flow qualification.

Executable steady and transient example

The executable TwoFluidPipeEvidenceExampleTest in src/test/java/neqsim/process/equipment/pipeline runs a 5 km, 0.30 m liquid-rich gas-condensate line at 50 kg/s, 60 bara and 50 degrees Celsius. It enables the opt-in shared slug force balance, interfacial pressure, and coupled pressure/momentum before steady initialization. The same object then advances ten one-second outer steps with unchanged boundaries.

The example fails unless:

Execute it from the repository root:

./mvnw -q -Dtest=TwoFluidPipeEvidenceExampleTest test

The exact Stage 5 execution values and source/head commit are recorded in the Stage 5 pull request and the issue #3298 series ledger. These values are reproducibility evidence for the example, not experimental qualification.

The prepared local execution on master 5bb230e1c2e2227074ce4d20ec19d6be91b53ec7 produced:

Quantity Result
Steady termination / iterations / tolerance CONVERGED / 16 / 1e-4
Steady outlet pressure 57.62005697 bara
Initial mean liquid holdup 0.269435835
Initial total inventory 72,424.765816 kg
Accepted transient duration 10.0 s
Maximum total-mass residual 1.92344e-11 kg
Maximum relative total-mass residual 2.65574e-16
Inventory drift 0.00977583%
Outlet-pressure drift 0.0%
Final mean liquid holdup 0.269473910

The Stage 5 pull request and issue ledger bind these values to the final evidence-only head.

Decision record

The five-stage evidence chain is recorded in issue #3298:

Open or draft work is not a released capability. Confirm the target release contains the referenced implementation before using its row.