Repository navigation
Description
Question. turbulence/README.md (cloud edges) finds single contour loops sitting about 0.035–0.039 above Kondev and Henley's (3 − H_eff)/2. It says the offset's "origin is open" and that it "is not explained". A published measurement bears directly on this. Should the README cite it, and say what it settles and what it does not?
This came out of the exploration of #39's avenues. Two independent checks confirmed it: a separate exact-surface reimplementation and a literature read.
The published measurement
Mandre and Kalda, Eur. Phys. J. B 84 (2011), arXiv:1105.2974, measured single-isoline dimensions on rough surfaces to about ±0.001, with box sizes 8 to 2048. Their table:
| H | D_h | Offset above (3 − H)/2 |
|---|---|---|
| 0.25 | 1.3958 ± 0.0014 | +0.021 |
| 0.5 | 1.2820 ± 0.0008 | +0.032 |
| 0.75 | 1.1565 ± 0.0010 | +0.032 |
| 0.875 | 1.0862 ± 0.0008 | +0.024 |
They conclude that the conjecture x_l ≡ 1/2 is "clearly rejected". Avellaneda et al. (1993) had already found 1.28 ± 0.015 at H = 0.5. Kondev, Henley and Salinas 2000 quote that result, and the README cites their paper.
An independent confirmation on exact surfaces
These are exact fractional Brownian surfaces, made by Stein circulant embedding, whose structure function is exactly r^(2H) at every separation. Fitting P against R_g over R_g = 8 to 256:
| H | Run 1 | Independent reimplementation | Offset above (3 − H)/2 |
|---|---|---|---|
| 0.3 | 1.375 | 1.3764 | +0.025 |
| 0.5 | 1.283 | 1.2842 | +0.033 |
| 0.7 | 1.189 | 1.1897 | +0.039 |
- Calibration. The same pipeline recovers 7/4 for percolation hulls (1.746 ± 0.004) and 3/2 for the Gaussian free field (1.498 ± 0.001).
- Scale dependence. At H = 0.3 and 0.5 the offset is flat within errors in two-octave windows from R_g = 4 to 256. At H = 0.5 it shows no bend-down out to R_g ≈ 512.
Caveats
-
Estimator bias at large H. The direct P-against-R_g fit overshoots there:
- at H = 0.9 it gives 1.105, which is on the rigorous bound 2 − H and about 0.035 above Mandre and Kalda;
- at H = 0.7 the local offset falls with scale, from +0.049 to +0.030.
So read values at H ≥ 0.7 as upper estimates. Kondev, Henley and Salinas's Table I shows the same overshoot of the direct estimator at α ≥ 0.6, and the README already notes this.
-
What the offset means is unresolved. Reading it as 2x_l ≈ 0.93 at H = 0.5 assumes hyperscaling and the sum rule. Two published results point the other way:
- published τ exponents (Kondev, Henley and Salinas's finite-size scaling) imply D at or below (3 − H)/2;
- their direct G(r) gives 2x_l ≈ 1.00.
Nobody has measured D, τ and G(r) on the same loops. That measurement would separate "x_l ≠ 1/2" from "hyperscaling or the sum rule fails here".
-
Coastlines cannot decide it. Ten real coastlines (arXiv:2409.20315) fit both (3 − H)/2 and the exact-fBm curve within their ±0.02 errors.
-
Mandre and Kalda's surfaces. They are sums of random cones, so only approximately Gaussian, and the authors chose the layering by hand. The agreement with exact-fBm numbers strengthens both measurements, but proves neither.
First steps
- Cite Mandre and Kalda 2011 beside Kondev and Henley 1995 in
turbulence/README.md, and replace "its origin is open" with what is known. The offset is measured, and at H ≈ 0.5 it agrees with the README's own. - Optionally, add an exact-fBm calibration to
contours.jsat H = 0.5, alongside the periodic Fourier surfaces, to separate the finite-grid H_eff effect from the intrinsic offset. - The open part is measuring τ and G(r) on the same loops. That could become a short study of its own.
Activity