Skip to content

Glossary

English | Japanese

« Home

Terms are arranged in 3 groups, in the order you meet them when reading a case page from top to bottom. The headwords stay in English because they are exactly the words that appear in the code, the logs and the section headings of the pages; only the explanations are translated. Each entry links the cross-cutting pages (Index / Models / Validation status) that can be reached from the definition.

Group 1 — Until the case is up and running

  • target / case — The build target orheo-<slug>. The convention is 1 case = a single .cpp in native/benchmarks/, so reading a case is the same as reading that 1 file. A CPU target and a -gpu twin target are built from the same source.
  • lattice (D2Q9 / D3Q19 / D3Q27 / D2Q5 / D2Q8) — The discrete velocity set. DxQy means x dimensions and y directions. Fluids mainly use D3Q19/D3Q27, scalar transport uses D2Q5, and radiative transport uses the discrete-ordinates D2Q8. It is fixed at compile time and cannot be changed at run time. Index lists all cases by lattice.
  • pseudo-2D — Many cases with "2d" in their name run not on D2Q9 but on D3Q19 with nz=1 and periodic z. This is the house style for validating exactly the same kernel as the 3D production runs on a 2D problem. It comes with a pitfall: if the force sum drops the links with c_z≠0, about 1/3 of the force is lost (the theory section of the cylinder2d note explains this in detail).
  • lattice units / u_lat — The non-dimensional unit system with dx = dt = 1. OpenRheo has no UnitConverter class; each benchmark chooses u_lat, N and Re directly, which fixes τ and Mach. Every velocity, viscosity and force printed in the logs is a value in this unit system.
  • τ (relaxation time) — The relaxation time of the collision. Since ν = c_s²(τ−1/2), in practice it is set as τ = 3·u_lat·N/Re + 0.5 (the header of ghia_re1000 contains this formula verbatim). 0.5 is the stability limit, and raising Re or lowering N slides τ toward it (Choosing units).
  • Mach number — u_lat / c_s = u_lat·√3. LBM is weakly compressible, so the smaller Ma is, the closer the result is to the incompressible solution. The default in many cases is u_lat ≈ 0.05 (Ma ≈ 0.087), but there are exceptions — ghia_re1000 and kbc_cavity2d use 0.1, and venturi3d uses 0.04 because the flow speeds up in the throat. If the peak Ma exceeds 0.1, suspect compressibility error.
  • module — 1 header in native/include/openrheo_amr/*.hpp. Cases are written as compositions of modules, and each module has its own golden ctest gate. Which case contains which module is listed in the mapping table on Models.
  • stepper / closure — The 2 kinds of module. A stepper owns the distribution functions and advances them in time (lbm_d3q19, two_phase, thermal_ddf). A closure is a pure function that only replaces coefficients (power-law viscosity, Darcy drag, EDL force — closures.hpp). Closures compose: a power-law fluid in a porous medium is written just by plugging 2 closures into the same stepper.
  • named config — The KEY=VALUE configuration file read by the forest cases (named_args.hpp). Dense-lattice cases take positional arguments, and the authority for them is the Usage line at the top of the source. If you do not know that the 2 styles coexist, it can look as if your arguments were silently ignored.
  • forest / block-forest AMR — A refined grid that arranges small fixed-size leaf blocks in an octree. The level is raised only where you want finer cells, and the grid is regridded to follow the flow. level, refinement band and sensor are the associated vocabulary; every case whose name starts with forest- runs on this machinery.
  • quick_args — The argument set shown in the Build & run section of a gallery page. It is a reduced version that prioritizes getting a picture within a few minutes; it is not the production or validation argument set. For the serious arguments, read the Usage line in the source header and its defaults.

Group 2 — During the time stepping

  • collision operator (BGK / regularized / TRT / MRT / central-moment / cumulant / KBC / RR / HRR) — The means of paying for stability. In the regime where plain BGK breaks down as τ→0.5, non-equilibrium regularization (regularized/RR/HRR), per-moment relaxation (MRT/CM), cumulants and entropic control (KBC) each extend the stable range in their own way. Some cases, such as kbc_cavity2d, switch between bgk|mrt|kbc on the same lattice and measure STABLE/DIVERGED. Which case uses what is in Models.
  • Smagorinsky / WALE / dynamic SGS — LES subgrid-stress models. They add the momentum transport of unresolved eddies as a local eddy viscosity ν_t and raise the effective τ — which as a result also acts as stabilization. Standard equipment in the turbulence and external-aerodynamics cases.
  • wall model — Wall treatment in LES that replaces the stress in the 1st cell at the wall with the log law or Spalding's law. Since full resolution down to the wall is no longer needed, the resolution of external-aerodynamics cases can be lowered by 1 whole level.
  • halfway bounce-back — A reflecting boundary that places the wall halfway between cells. If the streaming destination is solid, the distribution is simply reversed in place, giving 2nd-order-accurate no-slip. A moving wall adds a momentum term on reversal (the driven wall in couette and the lid of the cavity work this way).
  • Bouzidi — Interpolated bounce-back for walls that cut obliquely between cells. Curved boundaries in the forest cases use this (qlink). By contrast, the cylinder on the dense lattice is a staircase mask, and its Cd coming out a few % high is a known consequence of that (the cylinder2d note explains this with measurements).
  • Guo non-equilibrium extrapolation — The way inflow (prescribed velocity) and outflow (fixed pressure) boundaries are imposed. The distribution in a boundary cell is built as "the equilibrium of the prescribed macroscopic quantities + the non-equilibrium part extrapolated from the neighbouring cell". Almost every open boundary in this gallery uses it.
  • Guo forcing — The way body forces are applied (Guo 2002). The uniform force driving a channel, Darcy drag, the EDL force and Boussinesq buoyancy all enter the distribution functions through this path. What distinguishes it from naive addition is that it also accounts for the force shifting the very definition of velocity by half a step.
  • MEM (momentum-exchange method) — The way the force on a body is measured: count the momentum carried away by each link that bounces back from the wall and sum it up. Every Cd/Cl in this gallery is measured this way. sphere-drag is the minimal case that pins down its sign and magnitude against the Clift–Gauvin correlation.
  • fill fraction — The cell fill level in the free-surface VOF method (0 = gas, 1 = liquid). Mass is conserved while cells switch between the gas / interface / liquid types, and the interface in interface cells is reconstructed with PLIC. Core vocabulary of the dam-break and sloshing cases.
  • phase field (Allen–Cahn) — Conservative interface tracking for two-phase systems. The order parameter is transported by the Allen–Cahn equation and paired with a Fakhari-type high-density-ratio hydrodynamics solver. Unlike VOF, the interface is represented as a transition layer of finite thickness.
  • DDF (double distribution function) — A 2nd distribution function for thermal convection (thermal_ddf.hpp). Besides the distribution for the velocity field, it keeps a distribution for the temperature field, and the 2 are coupled both ways through Boussinesq buoyancy. The foundation of the thermal cases.
  • stdpar — ISO C++ standard parallel algorithms (std::execution::par_unseq). A single source is dispatched with nvc++ -stdpar=gpu / multicore, and the backend is decided at compile time. There is no run-time fallback, so "the GPU binary ran on the login node" proves nothing.

Group 3 — When reading the results

  • QoI / RESULT line — The validation quantities each case prints itself (L2 error, Cd, St, reattachment length…). Read these before the figures — that is the first rule for reading this gallery. They are the lines that decide "did it work" by numbers rather than by eye.
  • tier (T0–T3) — The validation rank. T0 = analytical solution + order of convergence, T1 = automatic gate with a reference, T2 = single-point comparison, T3 = demo. A tier displays "strength of evidence", not "correctness"; Validation status lists it for all cases.
  • gate (direct / module) — A golden ctest gate. direct runs the benchmark itself (currently only cavity3d_d3q27_ghia); module inherits the gate of the module behind it. It reflects the idea that in a compositional architecture the physics only needs to be validated 1 time, on the module side (Validating).
  • rel. err / L2 / EOC — The relative error against the reference, the L2 norm over the whole profile, and the order of convergence under grid refinement (experimental order of convergence). A single-point rel. err is the basis for T2; L2 + order is the basis for T0.
  • MLUPS — million lattice updates per second. A performance measure listed on Performance. That page states explicitly that it should not be used to compare cases with different lattices, physics or collision operators.
  • gallery run / run_id — The 1 recorded run that produced a page's figures, videos and logs (runs.json). Pages print the run date and commit, so you can always trace when, and with which code, a figure was made. Runs that were cut short say so.
  • source_sha / audit — Pins a note (the editorial sections of each page) to the commit of the .cpp it describes that was read when the note was written. When the source moves, wiki_pages.py audit reports the note as stale. This mechanism, which separates "facts that can be regenerated", "facts pinned to a run" and "prose pinned to the time of writing", is the basis for this wiki's reliability.
  • Known-issues — An honest page listing, with reasons, the cases that failed, produced no figure or have not been run. Being non-empty is normal, and the process of filling it in is itself a record of development. We recommend making it a habit to check this page and each page's Validation table before trusting a figure.