OpenLB and FluidX3D

Eddy3D’s two lattice-Boltzmann engines were scored on Cases A and B, on the same points as OpenFOAM 12; their scores and figures are on the Case A and Case B Validation pages. They are not scored on Cases C to E, which measure at 2 m. In empty-domain controls OpenLB’s ground wall model holds every point at that height within 5 % of the log law in none of the three: on openlb-wind:1.9.3-r6, whose ground wall is at \(z = 0\), its 2 m mean reads 0.82–0.86 of the log law in a narrow domain, and on the earlier 1.9.3-r4 and -r5 it was within 4 % in a wide one. FluidX3D’s ground wall function holds every point in all three; neither engine has been run on these cases. The measurements are below, and each case’s page says it for its own height. They have not been run on Cases F to M.

Contents
  1. OpenLB
  2. FluidX3D
  3. Why OpenLB is not scored on C to E
  4. Why FluidX3D is not yet scored on C to E

OpenLB

OpenLB (Eddy3D’s eddy3dWind: Smagorinsky, third-order regularised D3Q19, CPU image) solved the same two experiments with the same fitted inflow and probed the same measured points, at \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\): \(H/10\) for A, \(H/12.5\) for B, 1.3 and 2.6 million cells, 1.8 h and 4.1 h on four threads. \(H/20\) costs about 16 times the cell-steps on this image.

  • OpenLB meets the ground with a plain no-slip (bounce-back) wall and carries the roughness only in the inlet profile. In the undisturbed column at the windward face its \(U\) is 0.29 of the inlet log law one cell up, 0.69 at two, 0.95 at three and 1.02–1.07 from four. Case A’s lowest plane sits 0.6 cells up.
  • Eddy3D’s OpenLB has had an opt-in ground wall model since 2026-10-20: OpenLB’s wall-modelled boundary with a power law, Spalding’s law, or a rough log law with the inflow’s \(z_0\). The A and B runs predate it.
  • OpenLB’s STL reader segfaults when lattice nodes sit exactly on its octree planes, as in Case B at this spacing; Eddy3D moves such a lattice by \(10^{-4}\thinspace\mathrm{d}x\) (LBMCaseXml).

FluidX3D

FluidX3D (Eddy3D’s GPU lattice-Boltzmann engine: D3Q19 with Smagorinsky LES) solved both experiments at \(\mathrm{d}x = H/64\) (2.5 mm and 3.125 mm; 117 and 121 million cells) with the same fitted log-law inflow, averaging over 200 \(H/U_H\) after 40 of spin-up (A) and over 500 after 100 (B). Each run has an empty-domain control on the same lattice, and a point is scored only at a height where that control holds the imposed log law within 5 %: 116 of Case A’s 186 points and 88 of Case B’s 224.

  • In these runs FluidX3D meets the ground with a bounce-back wall and no roughness wall function. In the empty domain the flow at Case A’s lowest plane (\(z = 0.01\thinspace\mathrm{m}\), 4 cells up) reads 0.52 of the imposed log law on average, and 0.86 at \(z = 0.04\thinspace\mathrm{m}\); Case B’s three lowest heights read 0.77, 0.58 and 0.93. Those points are scored separately (velocity.*.excluded).
  • Eddy3D’s FluidX3D has had an opt-in ground wall function since 2026-11-05: the velocity three cells up gives, through the rough log law with the inflow’s \(z_0\), the shear stress the floor takes and the velocity the first fluid node is relaxed toward. The A and B runs predate it.

Why OpenLB is not scored on C to E

Cases C, D and E measure only at 2 m full scale (\(z = 0.1\thinspace D\) at 1:100, \(z/H = 0.02\) at 1:400, 2 m in the Niigata district): 1.25 cells above the ground at the A and B runs’ \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\). A point is scored only where every point at its height in an empty-domain control is within 5 % of the imposed log law, the bar FluidX3D’s points meet.

The first control: an empty domain 32 m wide and 80 m tall, \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\), each case’s own inflow at full scale (Case E: a log law fitted to its approach-flow profile below 60 m), the CPU image openlb-wind:1.9.3-r6, whose lattice puts its ground row half a cell under the domain so that 2 m is 2 m above the wall, read on three lines every 16 m where the case’s points would sit, after twice the time the 2 m layer takes to cross the domain and averaged over half of it. \(U\) at 2 m over the imposed log law, mean (range over the points), and the lowest height at which every point is within 5 %:

  Case C Case D Case E
distance from the inlet 80–320 m 770–1010 m 350–650 m
bounce-back 0.48 (0.37–0.64) 0.43 (0.16–0.66) 0.39 (0.26–0.58)
wall model, power law 0.93 (0.86–1.00) 1.01 (0.86–1.08) 0.83 (0.72–1.09)
wall model, rough law 0.85 (0.75–0.94) 0.86 (0.57–1.02) 0.82 (0.70–1.09)
lowest height with every point within 5 %, bounce-back 19.2 m (12 cells) 64 m (40 cells) 48 m (30 cells)
lowest height with every point within 5 %, power law 12.8 m (8 cells) none 32 m (20 cells)
lowest height with every point within 5 %, rough law 12.8 m (8 cells) none 48 m (30 cells)
Nomenclature: symbols, columns and scores

Geometry, reference scales and release

  • \(H\), \(b\), \(D\) (m) — height of the building, cube, tower or tree belt; width of the building; edge of the cubes of Case C: the lengths the cases are drawn and normalised by
  • \(U_H\) (m/s) — reference speed: the mean approach-flow speed at the height \(H\) (each Data page gives the value used)
  • \(U_R\), \(H_R\), \(Z_R\) (m/s, m) — reference speed and height of a workbook: Case E, \(U_R\) at \(Z_R\) = 250 m full scale (1 m in the model); Case M, \(U_R\) and \(H_R\) at the reference point of the tunnel and of each field run
  • \(U_{\mathrm{ref}}\), \(L_{\mathrm{ref}}\) (m/s, m) — reference speed and length of a case’s normalisation: \(U_H\) and \(H\) (\(U_R\) and \(H_R\) in Case M); for the velocity ratios the approach speed at 15.9 m (Case E) or at 500 m (Case F)
  • \(U_b\) (m/s) — Case G: mean approach-flow speed at 9 m, the normalisation of its approach-flow table
  • \(q\), \(c_{\mathrm{gas}}\) (m³/s, ppm) — tracer release of the experiment: volume flow of the released gas, and its tracer content (\(10^6\) ppm when pure); \(Q\) is the source strength given to a solve
  • \(R_b\) (dimensionless) — bulk Richardson number of the heated-floor cases J and L (negative: unstable stratification)
  • \(\mathrm{Re_H}\), \(\mathrm{Re_D}\) (dimensionless) — Reynolds numbers \(U_H H/\nu\) and \(U_H D/\nu\) (\(\nu\): kinematic viscosity of air)
  • \(u_\ast\), \(z_0\), \(\kappa\) (m/s, mm, dimensionless) — friction velocity and roughness length of the log law fitted to the approach flow, and the von Kármán constant (0.41)
  • \(\mathrm{WD}\) (column wind_direction_deg; °) — wind direction (each frame says how it is counted); \(\mathrm{WD_R}\) is the mean direction of a field run at its reference point (Case M)
  • \(x\), \(y\), \(z\) (columns x, y, z; m) — coordinates in the case’s frame (\(+x\), \(+y\) and \(+z\) are the positive directions; each Data page says which way \(+x\) points), \(z\) up from the floor of the tunnel or from the ground
  • \(\mathrm{d}x\) (m) — lattice spacing of the OpenLB and FluidX3D runs, in metres or as a fraction of \(H\) (\(H/64\))
  • \(L_r\) (m) — reattachment length behind the building, read from a wake line at ground level; reported as \(L_r/H\)

Velocity and turbulence

  • \(U\), \(V\), \(W\) (columns U, V, W; m/s) — mean velocity along \(+x\), \(+y\), \(+z\); \(V\) is positive toward \(+y\) in every case
  • \(\lvert\mathbf{U}\rvert\) (m/s) — magnitude of the mean velocity vector, \(\sqrt{U^2+V^2+W^2}\)
  • \(V_{\mathrm{s}}\) (column Vs; m/s) — mean omnidirectional scalar speed of a thermistor (never a vector magnitude)
  • \(\overline{u'u'}\), \(\overline{v'v'}\), \(\overline{w'w'}\) (columns uu, vv, ww; m²/s²) — velocity variances (overbar: time average, prime: fluctuation about it)
  • \(\overline{u'v'}\), \(\overline{u'w'}\), \(\overline{v'w'}\) (columns uv, uw, vw; m²/s²) — Reynolds shear stresses
  • \(k = \tfrac{1}{2}\left(\overline{u'u'}+\overline{v'v'}+\overline{w'w'}\right)\) (column k; m²/s²) — turbulent kinetic energy
  • \(U/U_{\mathrm{ref}}\) (column velocity_ratio; dimensionless) — mean speed over the case’s reference speed (\(V_{\mathrm{s}}/U_{\mathrm{ref}}\) for a thermistor)
  • \(\sigma_{U/U_{\mathrm{ref}}}\) (column velocity_ratio_sd; dimensionless) — standard deviation of velocity_ratio over the observations behind a field value (Case F)
  • \(k/U_{\mathrm{ref}}^2\) (column k_ratio; dimensionless) — turbulent kinetic energy over the square of the reference speed, where only normalised values are published (Case G)

Concentration and temperature

  • \(C\) (column C; ppm) — mean tracer-gas concentration as published (volume fraction \(\times 10^6\) of the released gas diluted in air)
  • \(c_{\mathrm{rms}}\) (column c_rms; ppm) — root mean square of the concentration fluctuation
  • \(c_0\) (ppm) — reference concentration, defined for each case on its Data page: \(c_0 = \dfrac{c_{\mathrm{gas}}\thinspace q}{U_H H^2}\) (\(U_R H_R^2\) in the denominator in Case M)
  • \(c^\ast\) (column concentration_ratio; dimensionless) — dimensionless concentration \(c^\ast = C/c_0\); the value of a solve is \(c\thinspace U_H H^2/Q\) for its mean concentration \(c\) and its source strength \(Q\)
  • \(\mathrm{Sc_t}\) (dimensionless) — turbulent Schmidt number of a solve (0.3, 0.5 and 0.7)
  • \(T\), \(T_{\mathrm{rms}}\) (columns T, T_rms; °C, K) — mean air temperature as published (Cases J and L) and the root mean square of its fluctuation
  • \(\overline{u'c'}\), \(\overline{v'c'}\), \(\overline{w'c'}\) (columns uc, vc, wc; ppm·m/s) — turbulent concentration fluxes
  • \(\overline{u'\theta'}\), \(\overline{v'\theta'}\), \(\overline{w'\theta'}\) (columns uT, vT, wT; K·m/s) — turbulent heat fluxes (\(\theta'\): the temperature fluctuation)
  • \(\theta = \dfrac{T_f - T}{T_f - T_H}\) (dimensionless) — floor-referenced temperature the solves of Cases J and L are scored on: 0 at the floor, 1 in the approach air at \(H\); \(T_f\) is the floor temperature and \(T_H\) the temperature of the approach flow at \(H\)

Scores

  • \(n\) — number of pairs of a measured value \(o_i\) and a modelled value \(p_i\) at one point
  • \(R\) — Pearson correlation coefficient of the pairs
  • \(R^2\) — square of \(R\): share of the measured variance a linear fit to the model explains
  • \(\mathrm{FAC2}\) — fraction of pairs with \(0.5 \le p_i/o_i \le 2\)
  • \(\mathrm{FB}\) — fractional bias \(2(\overline{o}-\overline{p})/(\overline{o}+\overline{p})\), positive when the model is low
  • \(\mathrm{HR}\) (hit rate) — fraction of pairs with \(\lvert p_i-o_i\rvert/\lvert o_i\rvert \le \delta_{\mathrm{rel}}\) or \(\lvert p_i-o_i\rvert \le \delta_{\mathrm{abs}}\), where \(\delta_{\mathrm{rel}} = 0.25\) and \(\delta_{\mathrm{abs}} = 0.05\thinspace U_H\) for a velocity component
  • \(\mathrm{NMSE} = \dfrac{\overline{(o-p)^2}}{\overline{o}\thinspace \overline{p}}\) — normalised mean square error
  • \(\mathrm{NAD} = \dfrac{\overline{\lvert o-p\rvert}}{\overline{o}+\overline{p}}\) — normalised absolute difference
  • \(\mathrm{MG} = \exp\overline{(\ln o-\ln p)}\) — geometric mean bias
  • \(\mathrm{VG} = \exp\overline{(\ln o-\ln p)^2}\) — geometric variance
  • The ratio of \(\mathrm{FAC2}\) keeps its sign, so a model of the other sign than the measurement is a miss; \(\mathrm{FB}\), \(\mathrm{NMSE}\) and \(\mathrm{NAD}\) divide by the means, so they are undefined (shown as —) for a signed component unless the measured and the modelled mean are both positive
  • VDI 3783-9 calls the hit rate \(q\) and the tolerances \(D\) and \(W\); on this site those letters keep their other meanings (\(q\) the tracer release rate, \(D\) the block size of Case C, \(W\) the vertical velocity)
  • The detection limit of a concentration is one value per case, \(10^{-3}\) of the largest measured value. It sets \(\delta_{\mathrm{abs}}\) of the hit rate and the both-below rule of \(\mathrm{FAC2}\), and floors the values in \(\mathrm{MG}\) and \(\mathrm{VG}\).
  • The scores \(\mathrm{FB}\), \(\mathrm{FAC2}\), \(\mathrm{NMSE}\), \(\mathrm{NAD}\), \(\mathrm{MG}\) and \(\mathrm{VG}\) follow Chang & Hanna (2004); the hit rate follows VDI 3783-9.

Quantities scored

  • \(U\), \(V\) and \(W\) (mean velocity components) over \(U_H\)
  • \(k\) (turbulent kinetic energy) over \(U_H^2\)
  • the speed \(\lvert\mathbf{U}\rvert\) and \(\sqrt{\lvert\mathbf{U}\rvert^2+2k}\), against a thermistor’s \(V_{\mathrm{s}}\)
  • \(c^\ast = C/c_0\) (dimensionless concentration) at turbulent Schmidt numbers \(\mathrm{Sc_t}\) of 0.3, 0.5 and 0.7
  • \(\theta\) (floor-referenced temperature)
  • On 1.9.3-r4, whose bounce-back wall sat half a cell (0.8 m) above \(z = 0\) and whose wall model’s first node a full cell up, the same control read 0.30, 0.26 and 0.23 with bounce-back, 1.00, 1.03 and 0.87 with the power law and 0.99, 1.01 and 0.94 with the rough law (Cases C, D, E); the lowest heights were the same, except Case E’s with the rough law, 32 m.
  • On 1.9.3-r4 the 2 m layer ran up to 6 % over the log law in the first 100 m with either wall law and 7–14 % under it 300–600 m in, in all three cases (in Case D the rough law was back within 4 % beyond 750 m), and beyond about 450 m from the inlet the control varied across its span at every height: at 12.8 m in Case D, 0.69–1.07 with bounce-back and 0.84–1.09 with the rough law.

The second control, on the CUDA builds of 1.9.3-r4 (Case E also on -r5), averaged over half a crossing, and of 1.9.3-r6, averaged over two: the same in a domain 256 m wide and 320 m tall read on seven lines, 1100 m long for Cases C and E and 1600 m for Case D (warmed up for 1.5 crossings), so that the outlet stays clear of the points:

  Case C Case D Case E
distance from the inlet 80–320 m 770–1010 m 350–650 m
1.9.3-r4, bounce-back 0.26 (0.05–0.44) 0.25 (0.18–0.32) 0.23 (0.11–0.34)
1.9.3-r4, wall model, rough law 1.04 (1.02–1.07) 1.00 (0.949–1.055) 1.03 (1.012–1.046)
1.9.3-r4, lowest height with every point within 5 %, bounce-back 19.2 m (12 cells) 19.2 m (12 cells) 32 m (20 cells)
1.9.3-r4, lowest height with every point within 5 %, rough law 2.4 m (1.5 cells) 8 m (5 cells) 2 m (1.25 cells)
1.9.3-r6, bounce-back 0.52 (0.05–0.77) 0.44 (0.38–0.51) 0.41 (0.31–0.51)
1.9.3-r6, wall model, rough law 0.88 (0.78–0.95) 0.86 (0.79–0.91) 0.90 (0.85–0.94)
1.9.3-r6, wall model, power law 0.96 (0.886–0.999) 0.98 (0.928–1.022) 0.90 (0.86–0.95)
1.9.3-r6, lowest height above 2 m with every point within 5 %, rough / power law 8 / 8 m 19.2 / 4.8 m 6.4 / 6.4 m
  • On 1.9.3-r6 (2026-11-09) no arm holds every 2 m point within 5 % over two crossings in any case. Case C’s power law does over the first half crossing (0.954–1.004) and not over two: its running mean falls from 0.980 to 0.961 as the averaging goes on. The bullets below are the 1.9.3-r4 and -r5 runs.
  • In this domain the rough law’s 2 m layer does not sink under the log law along the fetch: Case C’s reads 1.06 at the inlet and 0.99 at 750 m. Width and height changed together, so which of the two does it is not measured.
  • Case C: the 9 points over 5 % (of 112) all sit within 144 m of the inlet, where the layer is still coming down from its overshoot at the inlet (1.068 at 80 m); 160–400 m from the inlet every point is within 5 % (1.014–1.047).
  • Case E: every point is within 5 %, the highest (1.046) at the window’s upstream end, and 1.9.3-r5 reproduces that within 0.0004. Averaged over two crossings instead of half of one, five points are just over 5 % (0.985–1.0504): over the second crossing the 2 m mean was 1.008 against 1.028 over the first, and the points spread from 0.94 to 1.06 (32 of 133 over 5 %, 14 under).
  • Case D: 4 points over 5 % and 1 under (of 105; 0.949–1.055), on a pattern across the span that holds along the whole window and is symmetric about the centre line: the centre line reads 0.956, the lines 64 m either side 1.027. At any distance in the window the seven lines span 6.6–10.0 %. The outlet’s rise starts about 1450 m from the inlet.
  • A pressure outlet speeds the 2 m layer up over its last 260–380 m (0.8–1.2 domain heights): 400 m and 720 m long, as in the first control, this domain read up to 1.43 at Case C’s points and 1.25 at Case E’s.
  • Across the span the flow is symmetric about the centre line; at any distance in Case C’s and Case E’s windows the seven lines span at most 2.1 %.
  • The CUDA build (fp32) reproduces the CPU’s fp64 run of Case C’s narrow domain within \(2.1 \times 10^{-4}\) in \(U\) at all 1080 points.
  • On 1.9.3-r4 and -r5 OpenLB’s bounce-back wall sat half a cell (0.8 m) above \(z = 0\), where the inflow and the probes start, and the wall model’s wall at \(z = 0\) with its first node a full cell up; the bounce-back row of this table is read 1.2 m above its wall. 1.9.3-r6 (2026-11-06) moves the lattice’s ground row half a cell down, so both walls are at \(z = 0\) and the wall model’s first node is the half-cell node: in the first control the rough law’s 2 m mean goes from 0.94–1.01 to 0.82–0.86.
  • With 1.9.3-r4 the wall model took 6.7–6.9 times the bounce-back run’s wall time on 24 CPU threads and 3.6–4.3 times on one GPU; 1.9.3-r5 (2026-11-06) keeps its collision on the boundary row, and Case E’s wide control took 2.6 times on the same GPU (4268 s against 1611 s).
  • Scripts, probes, figures and run times: Eddy3D, testing/aij-openlb-ground-wall/.
Interpretation: OpenLB on Cases C to E
  • On 1.9.3-r4 and -r5 the wall model removes the collapsed floor: in the wide domain the 2 m mean goes from a quarter of the log law to within 4 % of it. On 1.9.3-r6, with the wall model’s first node half a cell up, it no longer does: over two crossings the wide domain’s 2 m mean reads 0.86–0.90 of the log law with the rough law and 0.90–0.98 with the power law.
  • On 1.9.3-r4 the narrow domain’s ±10 % along the fetch was the domain’s, not the ground’s.
  • An OpenLB run of these cases ends at least 1.2 domain heights past its last scored point, and its own control is the one that counts.
  • Case E: the half-crossing average met the bar, and nothing in it warned of the change: no point moved more than 0.003 over its second half. The layer changed after three crossings, so a control that can admit a 2 m point averages over several crossings after that, and so does the case run it admits.
  • Case C: over half a crossing, what keeps it unscored is the first 64 m of its window, where the wall-modelled layer is still settling from the inlet; whether the rest of the window holds over several crossings is not measured.
  • Case D: with the outlet clear of its points, what keeps it unscored is the pattern across the span; being symmetric about the centre, it is the 256 m domain’s, and a domain as wide as the case’s (about 1 km) is not measured.

Why FluidX3D is not yet scored on C to E

The OpenLB control above, on FluidX3D (Eddy3D’s pinned commit 8986874e, FP16S, an Apple M2 Max): open faces (slip sides and top, a pressure outlet) and a steady log-law inflow, the kind the A and B runs use, in the 32 m wide domain and in one 96 m wide. \(U\) at 2 m over the imposed log law, mean (range over the points):

  Case C Case D Case E
distance from the inlet 80–320 m 770–1010 m 350–650 m
bounce-back, 32 m wide 0.48 (0.38–0.60) 0.53 (0.47–0.60) 0.42 (0.30–0.52)
ground wall function, 32 m wide 1.00 (0.98–1.01) 1.02 (0.95–1.06) 0.98 (0.92–1.01)
bounce-back, 96 m wide 0.44 (0.38–0.55) 0.46 (0.20–0.74) 0.36 (0.20–0.56)
ground wall function, 96 m wide 1.00 (0.99–1.01) 0.96 (0.954–0.969) 0.99 (0.98–0.99)
Nomenclature: symbols, columns and scores

Geometry, reference scales and release

  • \(H\), \(b\), \(D\) (m) — height of the building, cube, tower or tree belt; width of the building; edge of the cubes of Case C: the lengths the cases are drawn and normalised by
  • \(U_H\) (m/s) — reference speed: the mean approach-flow speed at the height \(H\) (each Data page gives the value used)
  • \(U_R\), \(H_R\), \(Z_R\) (m/s, m) — reference speed and height of a workbook: Case E, \(U_R\) at \(Z_R\) = 250 m full scale (1 m in the model); Case M, \(U_R\) and \(H_R\) at the reference point of the tunnel and of each field run
  • \(U_{\mathrm{ref}}\), \(L_{\mathrm{ref}}\) (m/s, m) — reference speed and length of a case’s normalisation: \(U_H\) and \(H\) (\(U_R\) and \(H_R\) in Case M); for the velocity ratios the approach speed at 15.9 m (Case E) or at 500 m (Case F)
  • \(U_b\) (m/s) — Case G: mean approach-flow speed at 9 m, the normalisation of its approach-flow table
  • \(q\), \(c_{\mathrm{gas}}\) (m³/s, ppm) — tracer release of the experiment: volume flow of the released gas, and its tracer content (\(10^6\) ppm when pure); \(Q\) is the source strength given to a solve
  • \(R_b\) (dimensionless) — bulk Richardson number of the heated-floor cases J and L (negative: unstable stratification)
  • \(\mathrm{Re_H}\), \(\mathrm{Re_D}\) (dimensionless) — Reynolds numbers \(U_H H/\nu\) and \(U_H D/\nu\) (\(\nu\): kinematic viscosity of air)
  • \(u_\ast\), \(z_0\), \(\kappa\) (m/s, mm, dimensionless) — friction velocity and roughness length of the log law fitted to the approach flow, and the von Kármán constant (0.41)
  • \(\mathrm{WD}\) (column wind_direction_deg; °) — wind direction (each frame says how it is counted); \(\mathrm{WD_R}\) is the mean direction of a field run at its reference point (Case M)
  • \(x\), \(y\), \(z\) (columns x, y, z; m) — coordinates in the case’s frame (\(+x\), \(+y\) and \(+z\) are the positive directions; each Data page says which way \(+x\) points), \(z\) up from the floor of the tunnel or from the ground
  • \(\mathrm{d}x\) (m) — lattice spacing of the OpenLB and FluidX3D runs, in metres or as a fraction of \(H\) (\(H/64\))
  • \(L_r\) (m) — reattachment length behind the building, read from a wake line at ground level; reported as \(L_r/H\)

Velocity and turbulence

  • \(U\), \(V\), \(W\) (columns U, V, W; m/s) — mean velocity along \(+x\), \(+y\), \(+z\); \(V\) is positive toward \(+y\) in every case
  • \(\lvert\mathbf{U}\rvert\) (m/s) — magnitude of the mean velocity vector, \(\sqrt{U^2+V^2+W^2}\)
  • \(V_{\mathrm{s}}\) (column Vs; m/s) — mean omnidirectional scalar speed of a thermistor (never a vector magnitude)
  • \(\overline{u'u'}\), \(\overline{v'v'}\), \(\overline{w'w'}\) (columns uu, vv, ww; m²/s²) — velocity variances (overbar: time average, prime: fluctuation about it)
  • \(\overline{u'v'}\), \(\overline{u'w'}\), \(\overline{v'w'}\) (columns uv, uw, vw; m²/s²) — Reynolds shear stresses
  • \(k = \tfrac{1}{2}\left(\overline{u'u'}+\overline{v'v'}+\overline{w'w'}\right)\) (column k; m²/s²) — turbulent kinetic energy
  • \(U/U_{\mathrm{ref}}\) (column velocity_ratio; dimensionless) — mean speed over the case’s reference speed (\(V_{\mathrm{s}}/U_{\mathrm{ref}}\) for a thermistor)
  • \(\sigma_{U/U_{\mathrm{ref}}}\) (column velocity_ratio_sd; dimensionless) — standard deviation of velocity_ratio over the observations behind a field value (Case F)
  • \(k/U_{\mathrm{ref}}^2\) (column k_ratio; dimensionless) — turbulent kinetic energy over the square of the reference speed, where only normalised values are published (Case G)

Concentration and temperature

  • \(C\) (column C; ppm) — mean tracer-gas concentration as published (volume fraction \(\times 10^6\) of the released gas diluted in air)
  • \(c_{\mathrm{rms}}\) (column c_rms; ppm) — root mean square of the concentration fluctuation
  • \(c_0\) (ppm) — reference concentration, defined for each case on its Data page: \(c_0 = \dfrac{c_{\mathrm{gas}}\thinspace q}{U_H H^2}\) (\(U_R H_R^2\) in the denominator in Case M)
  • \(c^\ast\) (column concentration_ratio; dimensionless) — dimensionless concentration \(c^\ast = C/c_0\); the value of a solve is \(c\thinspace U_H H^2/Q\) for its mean concentration \(c\) and its source strength \(Q\)
  • \(\mathrm{Sc_t}\) (dimensionless) — turbulent Schmidt number of a solve (0.3, 0.5 and 0.7)
  • \(T\), \(T_{\mathrm{rms}}\) (columns T, T_rms; °C, K) — mean air temperature as published (Cases J and L) and the root mean square of its fluctuation
  • \(\overline{u'c'}\), \(\overline{v'c'}\), \(\overline{w'c'}\) (columns uc, vc, wc; ppm·m/s) — turbulent concentration fluxes
  • \(\overline{u'\theta'}\), \(\overline{v'\theta'}\), \(\overline{w'\theta'}\) (columns uT, vT, wT; K·m/s) — turbulent heat fluxes (\(\theta'\): the temperature fluctuation)
  • \(\theta = \dfrac{T_f - T}{T_f - T_H}\) (dimensionless) — floor-referenced temperature the solves of Cases J and L are scored on: 0 at the floor, 1 in the approach air at \(H\); \(T_f\) is the floor temperature and \(T_H\) the temperature of the approach flow at \(H\)

Scores

  • \(n\) — number of pairs of a measured value \(o_i\) and a modelled value \(p_i\) at one point
  • \(R\) — Pearson correlation coefficient of the pairs
  • \(R^2\) — square of \(R\): share of the measured variance a linear fit to the model explains
  • \(\mathrm{FAC2}\) — fraction of pairs with \(0.5 \le p_i/o_i \le 2\)
  • \(\mathrm{FB}\) — fractional bias \(2(\overline{o}-\overline{p})/(\overline{o}+\overline{p})\), positive when the model is low
  • \(\mathrm{HR}\) (hit rate) — fraction of pairs with \(\lvert p_i-o_i\rvert/\lvert o_i\rvert \le \delta_{\mathrm{rel}}\) or \(\lvert p_i-o_i\rvert \le \delta_{\mathrm{abs}}\), where \(\delta_{\mathrm{rel}} = 0.25\) and \(\delta_{\mathrm{abs}} = 0.05\thinspace U_H\) for a velocity component
  • \(\mathrm{NMSE} = \dfrac{\overline{(o-p)^2}}{\overline{o}\thinspace \overline{p}}\) — normalised mean square error
  • \(\mathrm{NAD} = \dfrac{\overline{\lvert o-p\rvert}}{\overline{o}+\overline{p}}\) — normalised absolute difference
  • \(\mathrm{MG} = \exp\overline{(\ln o-\ln p)}\) — geometric mean bias
  • \(\mathrm{VG} = \exp\overline{(\ln o-\ln p)^2}\) — geometric variance
  • The ratio of \(\mathrm{FAC2}\) keeps its sign, so a model of the other sign than the measurement is a miss; \(\mathrm{FB}\), \(\mathrm{NMSE}\) and \(\mathrm{NAD}\) divide by the means, so they are undefined (shown as —) for a signed component unless the measured and the modelled mean are both positive
  • VDI 3783-9 calls the hit rate \(q\) and the tolerances \(D\) and \(W\); on this site those letters keep their other meanings (\(q\) the tracer release rate, \(D\) the block size of Case C, \(W\) the vertical velocity)
  • The detection limit of a concentration is one value per case, \(10^{-3}\) of the largest measured value. It sets \(\delta_{\mathrm{abs}}\) of the hit rate and the both-below rule of \(\mathrm{FAC2}\), and floors the values in \(\mathrm{MG}\) and \(\mathrm{VG}\).
  • The scores \(\mathrm{FB}\), \(\mathrm{FAC2}\), \(\mathrm{NMSE}\), \(\mathrm{NAD}\), \(\mathrm{MG}\) and \(\mathrm{VG}\) follow Chang & Hanna (2004); the hit rate follows VDI 3783-9.

Quantities scored

  • \(U\), \(V\) and \(W\) (mean velocity components) over \(U_H\)
  • \(k\) (turbulent kinetic energy) over \(U_H^2\)
  • the speed \(\lvert\mathbf{U}\rvert\) and \(\sqrt{\lvert\mathbf{U}\rvert^2+2k}\), against a thermistor’s \(V_{\mathrm{s}}\)
  • \(c^\ast = C/c_0\) (dimensionless concentration) at turbulent Schmidt numbers \(\mathrm{Sc_t}\) of 0.3, 0.5 and 0.7
  • \(\theta\) (floor-referenced temperature)
  • With the ground wall function every 2 m point is within 5 % of the log law in all three cases in the 96 m wide domain. In the 32 m wide one only Case C’s are: the D and E points stray beyond about 450 m from the inlet, where the three lines drift apart, as OpenLB’s do in that domain.
  • With synthetic turbulence at the inlet, Eddy3D’s default for FluidX3D, the 2 m layer runs fast: 1.14 (1.12–1.18) and 1.18 (1.13–1.24) in Case D’s two domains, 1.05 (0.99–1.10) in Case C’s narrow one. Only Case E’s wide domain holds every point within 5 % (0.96–1.03).
  • The 2 m means at the averaging’s halfway point and at its end agree within 0.003 in every steady run. With the steady inflow the wall-function runs took 2–6 % longer than their bounce-back twins (whole runs of 17–76 s, the compile included).
  • The Niigata district fits on the GPU of the A and B runs at up to about 1.6 cells per 2 m (an estimate from the lattice sizes); \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\) is 1.25.
  • FluidX3D has not been run on Cases C to E.
  • Scripts, probes and run times: Eddy3D, testing/aij-fluidx3d-ground-wall/.
Interpretation: FluidX3D on Cases C to E
  • The ground wall function removes the collapsed floor: with the steady inflow the 2 m mean goes from 0.36–0.53 of the log law to 0.96–1.02.
  • With the steady inflow the ground no longer keeps these cases unscored; a run at \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\) with the wall function is scored wherever its own control holds the bar.
  • The 32 m domain’s drift in Cases D and E is the domain, not the ground: 96 m wide, the same ground holds.