Skip to content

settling-sphere3d

English | Japanese

« Home · « Particles and fluid-structure interaction

Fully-resolved settling-sphere FSI benchmark (Ten Cate et al. 2002). --------------------------------------------------------------------------- A single rigid sphere released from rest in a closed, quiescent container settles under gravity and is TWO-WAY coupled to the fluid: the fluid exerts a hydrodynamic force on the sphere (momentum exchange), the sphere's motion drives the fluid through a MOVING no-slip boundary. The signature result is the vertical-velocity history v(t): the sphere accelerates from rest, approaches a terminal velocity U_inf, then DECELERATES as it nears the bottom wall (the lubrication squeeze of the closing gap). Reproducing that accelerate -> plateau -> wall-approach-deceleration curve, and the terminal velocity, is the validation.

Geometry

settling-sphere3d geometry

Animation

Fields

The fields this computation wrote at its last output step, plotted as they are. 3D cases show the mid-plane.

settling-sphere3d umag

settling-sphere3d ux

Results

A fully resolved FSI reproduction of the PIV experiment of Ten Cate et al. (2002), case E3 (Re = 11.6, ρ_f = 962 kg/m³, μ = 0.113 Pa·s, experimental terminal velocity U_inf = 0.091 m/s, nylon sphere d = 15 mm, ρ_p/ρ_f = 1.1642). Grid 107×107×171 = 1.96 M cells, d = 16 cells (dx = 9.375e-4 m, dt = 2.060e-4 s), MRT (τ = 0.5828, ν_lat = 2.7586e-2), target lattice velocity 0.020 (Ma = 0.035), g_lat = 4.442e-4, initial height z0 = 120.2 mm. Wall-clock time 125.5 s on 72 Grace CPU cores.

The signature is the 3 phases of the vertical velocity history v(t) — acceleration → terminal plateau → deceleration on approaching the wall — which the CSV traces directly:

  • Acceleration: v/U_inf = 0.239 (t = 0.041 s) → 0.641 (0.206 s) → 0.810 (0.371 s).
  • Plateau: maximum v/U_inf = 0.9474 (t = 1.154 s, t* = 7.0, v = 0.08621 m/s). That is, the computed terminal velocity is 94.7% of the experimental 0.091 m/s, about 5% low.
  • Deceleration on approaching the wall: v/U_inf = 0.827 at gap = 4.61 mm (t = 1.442 s), 0.644 at gap = 1.78 mm, and 0.464 at the final gap = 0.94 mm (t = 1.500 s). Stops at gap ≤ 1 cell (step 7280 / maxsteps 20038). The deceleration due to lubrication squeeze in the closing gap appears qualitatively, within the limits of the grid resolution.

Candidates for the 5% shortfall are the staircase sphere surface at nd = 16 (effective-radius error), the finite container (the same 100×100×160 mm box as the experiment, but the grid is a discretization of it), and the treatment of forces with respect to added mass at a near-neutral density ratio; the header anticipates all of these as CAVEATS. Because the grid mapping is derived self-consistently from (nd, ulat, case), the terminal velocity is not an input but a pure prediction, and this 94.7% is that prediction's score.

Governing equations

The sphere is a two-way-coupled rigid body. Newton's equation (semi-implicit Euler, dt = 1 lattice step):

\[
m_p\frac{d\boldsymbol{v}_p}{dt}=\boldsymbol{F}_h+(\rho_p-\rho_f)V_p\,\boldsymbol{g},
\qquad V_p=\tfrac43\pi R^{3}
\]

The fluid is D3Q19 MRT; the static walls of the closed box use halfway bounce-back. The reflected distributions on the sphere surface carry the moving-wall momentum term:

\[
f_i(\boldsymbol{x}_f)=f^{*}_{\bar i}(\boldsymbol{x}_f)
+6\,w_i\,\rho_w\,(\boldsymbol{c}_i\cdot\boldsymbol{v}_p)
\]

(Ladd 1994; Aidun–Lu–Ding 1998). The fluid force is the momentum exchange on the sphere's boundary links, \(\boldsymbol{F}=\sum \boldsymbol{c}_i (f^{*}_i+f_{\bar i})\) — the same convention as the one pinned against Clift–Gauvin in sphere_drag.cpp. Cells vacated by the sphere are refilled with the equilibrium distribution at the neighbour-averaged density and the wall velocity v_p (a known delicate step). A density ratio of 1.16 is prone to added-mass oscillations, and a low-pass filter on the force (relax) is available, but this run uses relax = 1.00 (no filter).

Grid mapping: choosing nd and the target lattice terminal velocity ulat closes everything: dx = d/nd, dt = ulat·dx/U_inf, ν_lat = ulat·nd/Re, g_lat = g·dt²/dx. This is why velocities and times in physical units appear directly in the CSV.

Code walkthrough

Source (440 lines).

%%{init: {'themeVariables': {'fontSize': '11px'}}}%%
flowchart TB
    subgraph BOX["Closed box 107×107×171 (100×100×160 mm)<br/>halfway BB on all six faces / interior MRT"]
      S(("Sphere d=16 cells<br/>moving-wall BB + MEM force<br/>covered/uncovered refill"))
      W["Bottom wall: on approach,<br/>deceleration by lubrication squeeze"]
      S -->|"gravity (ρp−ρf)Vg"| W
    end
  1. Mapping — derive and print (dx, dt, ν_lat, g_lat) from the physical constants of case = E3.
  2. collide/stream — MRT. The sphere mask is rebuilt every step from the sphere-centre position.
  3. Force — MEM sum over the boundary links. Optional low-pass filter (passed through in this run).
  4. Rigid-body update — Newton in 1 direction (vertical). Covered cells are removed, uncovered cells are refilled with equilibrium.
  5. Output — z, gap, v, F in physical units to the CSV every 200 steps. At gap ≤ 1 cell it prints "sphere reached the bottom wall; stopping." and exits.

Build & run

cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-settling-sphere3d

Cost of the computation that made the figures on this page (Miyabi-G Grace CPU (72 cores, -stdpar=multicore)): 125.5 s wall.

Things to try

  • Change nd from 16 → 24 (memory and time ~(24/16)⁴ ≈ 5 times). If the 5% shortfall in terminal velocity is a resolution error of the staircase sphere surface, it should shrink here.
  • Sweep case over 1–4. Trace the whole experimental series from E1 (Re = 1.5) to E4 (Re = 31.9) with the same code, and look at how the speed of reaching terminal velocity and the sharpness of the wall deceleration differ with Re.
  • Switch to collision = 1 (BGK). How different is v(t) between BGK and MRT at τ = 0.583? If the difference is small, the collision operator is not the dominant factor at this Re.
  • Try relax < 1. How does the filter change the added-mass oscillations at the near-neutral density ratio (the fine wiggles early in v(t))? Also confirm that the static terminal value does not move.

Validation

Tier T2 — single-point comparison
QoI terminal_velocity
Reference Ten Cate et al. (2002) Phys.Fluids 14:4012 settling sphere
Gate 🧩 module gate (rigid_body)

Source & run details

View source

Item Value
Target orheo-settling-sphere3d
Source native/benchmarks/settling_sphere3d.cpp
Lattice D3Q19
GPU twin yes (orheo-settling-sphere3d-gpu)
Modules lbm_d3q19.hpp
Gallery run 2026-07-27 @ 7628c0cf
Args ``
Wall time 125.48 s
CPU time 6632.55 s
Device Miyabi-G Grace CPU (72 cores, -stdpar=multicore)

Computation log

The actual output of this computation (head and tail).

Ten Cate settling sphere: case E3  Re=11.6  rho_f=962 mu=0.113 U_inf=0.091 m/s
  grid 107x107x171 = 1.96M cells,  d=16 cells (R=8.0),  dx=9.375e-04 m  dt=2.060e-04 s
  nu_lat=2.7586e-02 tau=0.5828 (MRT)  ulat_target=0.020 (Ma=0.035)  g_lat=4.442e-04  rho_p/rho_f=1.1642
  z0=128.2 cells (120.2 mm), maxsteps=20038, force relax=1.00
step      0  t=0.000s  z=120.23mm gap=112.73mm  vz=0.0003 m/s (v/Uinf=0.003)  Fz=1.421e-14
step    200  t=0.041s  z=119.72mm gap=112.22mm  vz=0.0218 m/s (v/Uinf=0.239)  Fz=1.111e-01
step    400  t=0.082s  z=118.54mm gap=111.04mm  vz=0.0350 m/s (v/Uinf=0.385)  Fz=1.249e-01
step    600  t=0.124s  z=116.89mm gap=109.39mm  vz=0.0446 m/s (v/Uinf=0.490)  Fz=1.369e-01
step    800  t=0.165s  z=114.89mm gap=107.39mm  vz=0.0523 m/s (v/Uinf=0.574)  Fz=1.354e-01
step   1000  t=0.206s  z=112.60mm gap=105.10mm  vz=0.0584 m/s (v/Uinf=0.641)  Fz=1.393e-01
step   1200  t=0.247s  z=110.09mm gap=102.59mm  vz=0.0634 m/s (v/Uinf=0.697)  Fz=1.412e-01
step   1400  t=0.288s  z=107.39mm gap=99.89mm  vz=0.0675 m/s (v/Uinf=0.742)  Fz=1.414e-01
step   1600  t=0.330s  z=104.53mm gap=97.03mm  vz=0.0709 m/s (v/Uinf=0.779)  Fz=1.396e-01
step   1800  t=0.371s  z=101.55mm gap=94.05mm  vz=0.0737 m/s (v/Uinf=0.810)  Fz=1.507e-01
...

step   6200  t=1.277s  z=25.77mm gap=18.27mm  vz=0.0857 m/s (v/Uinf=0.942)  Fz=1.585e-01
step   6400  t=1.319s  z=22.24mm gap=14.74mm  vz=0.0851 m/s (v/Uinf=0.935)  Fz=1.522e-01
step   6600  t=1.360s  z=18.76mm gap=11.26mm  vz=0.0839 m/s (v/Uinf=0.922)  Fz=1.515e-01
step   6800  t=1.401s  z=15.35mm gap=7.85mm  vz=0.0814 m/s (v/Uinf=0.894)  Fz=1.661e-01
step   7000  t=1.442s  z=12.11mm gap=4.61mm  vz=0.0753 m/s (v/Uinf=0.827)  Fz=1.888e-01
step   7200  t=1.484s  z=9.28mm gap=1.78mm  vz=0.0586 m/s (v/Uinf=0.644)  Fz=2.554e-01
step   7280  t=1.500s  z=8.44mm gap=0.94mm  vz=0.0422 m/s (v/Uinf=0.464)  Fz=2.661e-01
sphere reached the bottom wall (gap<=1 cell); stopping.
wrote settling_sphere.csv

References

  • A. ten Cate, C. H. Nieuwstad, J. J. Derksen & H. E. A. van den Akker, "Particle imaging velocimetry experiments and lattice-Boltzmann simulations on a single sphere settling under gravity", Phys. Fluids 14(11) (2002) 4012–4025.
  • A. J. C. Ladd, J. Fluid Mech. 271 (1994), Parts I & II.
  • C. K. Aidun, Y. Lu & E.-J. Ding, J. Fluid Mech. 373 (1998).