Save a clean animation frame without ticks or labels. Creates a clean image suitable for animation with: - Black background (matchin

Run benchmarks for all three implementations.

OPTIMIZED fused streaming and bounce-back operation. Uses unrolled directions to eliminate if/elif branching for better performance.

Perform the streaming step of the LBM algorithm. Particles move along their velocity directions to neighboring lattice sites.

Viridis (piecewise linear like your Fortran) */

Visualize the flow field by plotting vorticity and streamlines. Saves the plot as an image file. When NOTICKS is True, creates clean

Visualize the flow field by plotting vorticity and streamlines. Saves the plot as an image file. When NOTICKS is True, creates clean

============================== RNG ============================== */ Simple xorshift32 per-call seeded engine -> reproducible per seed */

Initialize the benchmarker. Parameters: ----------- script_name : str Base name of the scripts t

Detect outliers using a simple percentage threshold relative to the mean. This is appropriate for performance benchmarking where we w

Apply velocity boundary conditions at inlet and outlet. Maintains constant velocity at domain boundaries.

Apply periodic boundary conditions to the distribution functions. This ensures particles leaving one edge of the domain re-enter from the opp

Perform the collision step using the BGK (Bhatnagar-Gross-Krook) approximation. Distribution functions relax toward their equilibrium values.

Compute macroscopic fluid variables (density and velocity) from the distribution functions using moment integrals.

Handle the no-slip boundary condition at the cylinder surface. Implements the bounce-back method where particles hitting the cylinder rev

Handle the no-slip boundary condition at the cylinder surface using CORRECTED bounce-back. OPTIMIZED VERSION: - Uses precomputed mas

Fused leapfrog integration step to eliminate Python ⇄ NumPy overhead. Implements the complete leapfrog scheme in a single JIT-compiled f

Perform the streaming step of the LBM algorithm. Particles move along their velocity directions to neighboring lattice sites.

dst[y,x,i] = src[y - CY[i], x - CX[i], i] with periodic wrap, no temp arrays. This avoids np.roll allocations which create full plane copies