24 simulation families, all of them. A portfolio that only shows the wins hides the judgement that matters, so the negative, unfinished and blocked work is listed with the same detail.
Forced-convection heat transfer, axisymmetric RANS, k-omega SST vs SST-LM transition
Wall-jet Nu error, r/D >= 2.5 (SST)
-4.3 to +0.4 %
Stagnation Nu error (SST)
+13 %
Mean |error| SST / SST-LM
12.3 % / 12.2 %
Limits. Single grid, no GCI. Plate heat flux is imposed (Nu is non-dimensional); experimental Nu uncertainty unknown. Non-precursor pipe inlet. SST ran 4000 iterations against 8000 for SST-LM and its wall jet beyond r/D 3 was still moving about 2 % per 500 iterations, so the 12.3 vs 12.2 % tie is unresolved. SST over-predicts the wall-jet velocity peak by 14-20 % at r/D 2.5-3. Neither model reproduces the secondary Nu peak; the negative result is kept. Data: ERCOFTAC C.25, CC BY-NC-SA 4.0.
Limits. A comparison, not an ASME V&V 20 validation: the experimental mean Nu 5.85 is quoted second-hand from the ERCOFTAC KBwiki UFR 4-10 page, with its definition and uncertainty unknown. 2-D slice of a 3-D rig; top and bottom wall temperatures are imposed from the experiment. The 20 % deficit appears in the wall layers (experimental gradient about 40 % steeper at mid-height, a 3-point inference); the 'too laminar' hypothesis is untested. Order 1.34 on 4 grids, LSR band 0.5 %. Data CC BY-NC-SA 4.0.
Steady laminar conjugate heat transfer, chtMultiRegionSimpleFoam, one unit cell (water, copper, cover, air pockets), four grids 11,660 to 268,522 cells
Model minus measured, TC1 to TC4, finest grid (Re 890, 100 W/cm2)
+0.95 / +2.09 / +1.87 / +1.42 C
Numerical uncertainty at TC1 (4-grid least squares, 95 %, this scheme only)
about 0.5 K
Measured TC1 (digitised from Fig. 7, +-0.5 C estimate)
40.13 C
Limits. A comparison, not a validation. The measured temperatures are digitised from a figure (+-0.5 C digitising estimate, kept separate from the paper's < 0.3 C thermocouple uncertainty) and only one operating point has a grid ladder. Only TC1 has a numerical band (about 0.5 K, 95 %, linearUpwind only); TC2 shows no grid convergence and TC3, TC4 and the outlet temperature get no band, because the coarsest grid is not a geometric refinement of the other three. The energy scheme undershoots the inlet temperature by up to 2.25 K on every grid, and a first-order run moved the thermocouples by 0.3-1.9 C, more than the TC1 band. Air-pocket geometry and the 467 um cell width are read from drawings; a unit cell has no plenum maldistribution. Paper copyright Elsevier; its figure is not reproduced.
NACA 4412 Cl vs experiment (post-hoc re-run, not blind)
+2.13 % (single 48k grid, Cl 1.5871 vs 1.554)
NACA 0012 Cl(h->0) vs CFL3D, GCI 1.00 %
+2.1 % (vs CFL3D, compressible M 0.15; +3.4-3.8 % vs Ladson)
Limits. An earlier GCI (p 1.91, 0.58 %) was wrong and is retracted; corrected p 1.24, band 1.0 %, 3 grids. The finest 0012 grid fails checkMesh on aspect ratio (1457, 12 cells, undocumented) and fine-grid force drift exceeds the freeze's own 1e-5 gate. Cd is not grid-converged. The 4412 figure is a re-run with a changed recipe after the experiment was known; the genuinely blind run (+6.4 %, band-pass) used a recipe later found defective and is invalid. The 4412 aft-upper Cp plateau fails. The experimental Cl 1.554 is not stored locally.
External aero, simpleFoam k-omega SST, 3D half model, 2.06M to 6.42M cells
Measured Cd, finest grid (6.42M), vs experiment
0.2856 vs 0.3070 (-7.0 %, inlet q)
Probe-q 3-grid extrapolation vs experiment
-16.7 % (vs -9.9 % with inlet q)
Limits. A fourth, coarser grid (1.2M cells, Cd 0.29998, solved after the write-up) breaks the convergence pattern (successive Cd differences 0.0001, 0.0087, 0.0057 do not shrink steadily): the 4-grid least-squares fit has no positive order, so the earlier Richardson value -9.9 % and its band [-18.8 %, -1.1 %] are withdrawn. The tunnel's reference dynamic pressure is unresolved: probe-normalised the same solves read about 2.5 % lower (3-grid extrapolations -9.9 % inlet vs -16.7 % probe). Lift is far off: -0.142 against -0.030 (the closure memo calls it a sign-magnitude failure and leaves it unresolved). The 1.2M grid had one skewness failure waived. Earlier -0.7 % and -5.1 % figures are superseded.
+25 to +26 % measured at equal grid (extrapolated pair +30 %) vs +33.5 %
Limits. Three grids with observed order 5-7, noise-limited (the medium-to-fine step equals the limit-cycle amplitude); no checkMesh log exists for the 5.59M grid. Yaw cases are single full-model grids; 0 deg is a half model. Wheels static, contact patch bridged, probe-q would read about -14.5 %. Lift sign flip is reproduced. Single experiment, so a comparison, not a validation.
Limits. The thrust match is not a precision claim: at J 0.523 a 1 % change in J moves CT by about 5 %, UIUC's uncertainty is unpublished, and the hub is estimated (blades-only CT +8 %). The loft departs 5 % from the published twist table. checkMesh failed on 2 blade faces (waived, recorded); about 11 of 15 layers achieved; blade y+ max 72.7; aft-chord first cells 16-107 um, unexplained after two trials. Power +11.6 % is unexplained and parked. 1000 iterations, single grid, no GCI.
35 deg GCI, realizable k-eps, separate 0.39-1.79M family (y+ 290-460, no layers)
~27 %
Limits. Pre-gate July study, wall functions, 1500 fixed iterations. The 27 % GCI belongs to a coarser mesh family whose 0.92M grid gives Cd 0.358 for the same model that reads 0.312 on the 1.1M sweep mesh: mesh spread exceeds the model gap. SST sits on the attached branch at 35 deg (+49 %). URANS 35 deg is +14 % vs Meile 0.279 (+25 % only against the Re-mismatched Ahmed 1984 value).
Isolated rotating wheel (Fackrell)
Negative
Wheel + ground, MRF hub, RANS and URANS, 3.9M to 9.3M cells
Fine grid Cd / Cl vs experiment
+9.0 % / +7.3 %
Limits. The medium-grid agreement (Cl -1.9 %, Cd +4.1 %) did not survive refinement: it was cancellation. Steady RANS is falsified for this wake.
pimpleFoam laminar 2D, O-grid ladder 5.6k to 45k cells
Strouhal number, simulation
0.1839 (finest grid; LSR 0.1828 +/- 0.97 %)
Strouhal number, reference
0.1841 (Qu 2013)
Limits. Only St is defensible: -0.13 % vs Qu et al. 2013 on the finest grid and within 0.9 % on all four, but the three finest grids are not monotone (GCI refused) and the extrapolated value is 0.1828. Cd +1.4 %, CL' +5.5 %, CD' +9.7 % with BAD fits (bands 9-18 %). The Curle noise level has no numeric reference; DNS St 0.183 is second-hand. A finer grid (89k) is owed.
solidDisplacementFoam / solids4foam, solid only (no fluid-structure coupling run)
Cook's membrane (plane strain) corner displacement, LSR of 4 grids
32.23 +/- 0.26 mm
Limits. Code-to-code comparison, not validation: plane-strain Cook variant (E 70 MPa, nu 1/3) against CalculiX, Code_Aster and Elmer; cantilever against a self-derived beam-theory value. The 20x20 mesh is 3.3 % below the solids4foam README, attributed by elimination to coarse-mesh code sensitivity. Hand-computed orders were retracted and replaced by the shared GCI tool; a false-convergence incident on the tutorial ladders was found and the runs redone.
Method of manufactured solutions (laminar Navier-Stokes)
Verified
Code verification of the solver setup
Observed order U / p
1.995 / 1.952
Limits. Laminar steady Navier-Stokes at Re about 50 on a unit square; says nothing about turbulence or walls. A k-omega SST manufactured solution now runs (N 20-320): the pressure order 2.004 is verified, the U/k/omega orders (2.4-2.6) are pre-asymptotic and not verified. Two seeded defects were caught.
Limits. References are XFOIL and TMR CFL3D, not experiment. Cd GCI is not converged. NACA 0015 Cd is +41 to +96 % and the 20 deg case did not converge.
Free-surface foil (Pernod / Duncan)
Caveated
interFoam VOF 2D, 91k cells
Cl_hydro and Cd at h/c 1.03
+7 % and -16 %
Limits. Earlier runs drifted by 23 to 38 %. No GCI.
Cavitation: Kermeen hydrofoil and Clark-Y
Negative
interPhaseChangeFoam, Schnerr-Sauer
Kermeen cavitating Cd
+103 to +116 %
Limits. The non-cavitating match needed a trip tuned to the experiment. Cavitating Cd is far off. The Clark-Y ladder is non-monotone and the finest grid blew up. The S1 ladder is invalid and its earlier finding was retracted for a Courant-number error.
Hull and strut free-surface resistance
Unfinished
3D VOF interFoam, DTC and DSYHS hulls
DTC resistance at step 1500 of 4000
+5.6 to +6.5 %
Limits. Not flat, no grid study; the reference came from a secondary transcription. DSYHS has no resistance number. The strut run was stopped.
Steady simpleFoam k-omega SST with wall functions, 75 % Ahmed 25 deg, isolated (651,166 cells) and two in line at 8 gaps (0.96M to 1.12M cells)
Isolated 25 deg Cd vs Pagliarella 0.30 (same scale and Re) / Meile 0.299 (full scale)
0.2798 (-6.7 % / -6.4 %)
Lead drag ratio at x/L 0.5 vs experiment (start- and mesh-dependent: 0.495 to 0.557)
0.555 vs 0.628 (-12 %)
Four-grid ladder at x/L 0.5 (0.19M to 6.96M cells)
did not converge; no numerical uncertainty
Limits. A comparison, not a validation, on one mesh per case. The isolated Cd is 6-7 % below two experiments; the like-for-like one is Pagliarella's 0.30 (same scale and Reynolds number), Meile's 0.299 is full scale at Re 2.78e6 in an open jet. The tandem trend is reproduced, but the lead body is 12-28 % low at close gaps (cause open) and the trail 4-11 % high at several gaps, partly because every ratio is divided by my own isolated Cd. Simulation bars are iterative scatter; the x/L 0.5 point moves 0.495-0.557 (lead) with start and mesh. The four-grid ladder did not converge (y+ drift 27 % / 26 %, non-similar refinement), so no band is quoted. Extended checkMesh fails 3 checks on every mesh. Stings, wall boundary layers and flow angularity not modelled; the experimental series is digitised from a thesis figure, licence not stated. The 35/25 cases are excluded (wrong branch, mixed normaliser).
Limits. Non-monotone: no valid GCI, no reference data. The finest run never finished.
Curved-flow wing (SRF)
Negative
SRF simpleFoam, 2D then 3D
Limits. Signs agree with Keogh 2014, magnitudes do not. The symmetry-floor gate failed and the endplate split is inverted; magnitudes are not trustworthy.
30P-30N multi-element airfoil
Unfinished
2D SST, structured grids
Limits. Cl 2.824 +/- 0.003 but the GCI is non-monotone (limit cycle) and the extension rung is paused.
DLR-F11 high-lift (HLPW-2)
Blocked
3D high-lift configuration
Limits. No result: the committee grid fails checkMesh (aspect ratio 175,108; non-orthogonality 89.99).