Makarov Physics Suite · published · build 24957107
The wing, the sphere, the plate and the other bodies now run under five different treatments of the scales the grid cannot resolve, switchable while the flow is running:
None (DNS on this grid) — the equations as they are, nothing added.
LES · Smagorinsky — the classic sub-grid eddy viscosity; watch it over-damp the boundary layer on the wing.
LES · WALE — the wall-adapting model that vanishes in laminar shear, so the separation belongs to the flow.
URANS · k-ω SST — the industry-standard averaged model: the wake becomes a smooth ribbon and k shows where the turbulence is.
DDES — the hybrid: SST inside the boundary layer, LES away from the wall. A new chart plots the drag and lift coefficients over time. Switching the model freezes the previous curves, so you can compare answers on the same body — they differ, and that is the point. Tune C_s, C_w and the inflow turbulence in the advanced panel.
LBM 2D gets Smagorinsky (computed from the non-equilibrium stress, the lattice-native way) and WALE closures; LBM 3D gets Smagorinsky. Both let the lattice hold Reynolds numbers the bare BGK collision could not. All of this runs on the CPU, CUDA and OpenCL backends alike, and every model is covered by the self-test suite.
The LBM 3D description used to mention an LES that was not there. Now it is.
LBM 3D: at low relaxation with a strong drive the lattice could blow up, the tracers left the box and Stop did not help. The driving force is now capped where the lattice cannot hold it (the panel says so), and a blown field restarts itself instead of showing NaN.
Cloth: a piece torn free on all sides stayed hanging in the air until touched. Torn edges now take the diagonal and bending links with them, and the piece falls.