README
======

This directory contains two 2D decaying turbulence datasets, each
representing a time-varying 2D vector (velocity) field.

Dataset 1: 1024x1024x1501Re10k
  Files: u_1024x1024x1501Re10k.f32, v_1024x1024x1501Re10k.f32
  Grid:       1024 x 1024
  Timesteps:  1501
  Per-frame size:  1024 x 1024 x 4 bytes = 4,194,304 bytes (4 MB)
  Total file size: 1501 x 4 MB = 6,295,650,304 bytes (~6004 MB / ~5.86 GB) per component

Dataset 2: 2048x2048x1001Re10k
  Files: u_2048x2048x1001Re10k.f32, v_2048x2048x1001Re10k.f32
  Grid:       2048 x 2048
  Timesteps:  1001
  Per-frame size:  2048 x 2048 x 4 bytes = 16,777,216 bytes (16 MB)
  Total file size: 1001 x 16 MB = 16,793,993,216 bytes (~16016 MB / ~15.64 GB) per component

Why two files per dataset (u and v):
  Each dataset is a 2D vector field, so at every grid point there are
  two velocity components: u (x-direction) and v (y-direction). These
  two scalar components are stored as two separate raw binary files
  rather than interleaved into a single file.

File format:
  Raw binary, single-precision float (float32), little-endian, no
  header. Frames are stored consecutively (one 1024x1024 or 2048x2048
  slice per timestep, concatenated over all timesteps). Dimensions are
  encoded in the file name as WIDTHxHEIGHTxTIMESTEPS.

  Note: the first timestep (t=0) of every file is an all-zero
  placeholder frame; the actual random band-limited initial condition
  begins at the second timestep (t=1). See the corresponding
  metadata_*.txt files for detailed per-file statistics (min, max,
  mean, variance, etc.).

How the data was generated:
  Both datasets are simulations of 2D decaying turbulence at Reynolds
  number Re = 10,000, generated using the Basilisk solver, with the
  initial velocity field created from a random band-limited
  initialization.

Source:
  [1] Basilisk solver: http://basilisk.fr/
  [2] M. Xia, Y. Li, P. Jiao, B. Wang, X. Liang and H. Guo, "Time-Varying
      Vector Field Compression with Preserved Critical Point
      Trajectories," 2026 IEEE 42nd International Conference on Data
      Engineering (ICDE), Montreal, QC, Canada, 2026, pp. 2235-2249,
      doi: 10.1109/ICDE65706.2026.00168.