`shear_flow`: shipped `dimensions/y` spans [0, 1] but the README (and the field data) say the domain is [-1, 1]

[arturtoshev]

The shear_flow README states:

Spatial domain size: $0\leq x \leq 1$ horizontally, and $-1 \leq y \leq 1$ vertically.

The shipped coordinate disagrees: dimensions/y is 512 points spanning 0 → 1, so the recorded wall-normal extent is half the documented one, and the origin is shifted.

Reproduction

import h5py, numpy as np

with h5py.File("shear_flow_Reynolds_1e5_Schmidt_2e0.hdf5") as f:
    x, y = f["dimensions/x"][:], f["dimensions/y"][:]

print(x.size, x.min(), x.max(), np.diff(x)[0])   # 256, 0.0, 1.0, 1/255
print(y.size, y.min(), y.max(), np.diff(y)[0])   # 512, 0.0, 1.0, 1/511  <- documented as [-1, 1]

The field data agrees with the README, not with the coordinate

The domain is doubly periodic and the flow is incompressible, so ∂ₓvₓ + ∂_y v_y = 0 exactly and both derivatives can be computed spectrally.

Using the recorded extents, on shear_flow_Reynolds_1e4_Schmidt_1e-1, frame 100:

test	with Ly = 1 (recorded)	with Ly = 2 (documented)
relative divergence residual	0.333	0.0041
slope of ∂_y v_y on ∂ₓvₓ	−1.99964 (corr −0.99989)	−1.0
dy / dx	0.499	0.998

Incompressibility demands a slope of exactly −1. Measuring −1.99964 at correlation 0.99989 says dy is understated by exactly a factor of 2, and correcting it also makes the grid isotropic, which a 512×256 array on a 2:1 domain should be. Reproduced on a second file (Reynolds_1e5_Schmidt_1e0, 0.4473 → 0.0108).

Note the divergence test constrains only the spacing, so it cannot see the origin on its own — the README is what establishes that the corrected coordinate is y = 2·y_shipped − 1, i.e. [-1, 1], rather than [0, 2].

Impact

Any consumer taking a wall-normal derivative from the recorded coordinate is off by 2×, and any consumer using absolute y positions is additionally offset by 1.

This may already have cost someone: issue #62 reported implausibly large PDE residuals on shear_flow, and the discussion attributed it entirely to a finite-difference vs pseudospectral mismatch before closing. A 2× error in dy produces exactly that symptom, and the coordinate was never examined.

Suggested fix

Ship dimensions/y as 512 points spanning [-1, 1] (dy = 2/511), matching the README. dimensions/x is already correct.

Checked before filing

Swept all open and closed issues and all merged PRs. Nothing mentions the shear_flow domain size. PR #13 ("Fix shear_flow README.md") touched the resolution and initial-condition description but not the domain line, and issue #62 is the near-miss described above.