Multi-Scale Wasserstein

Scale coherence, distributional drift, self-similarity
scaledim distribution cascade5 metrics

What It Measures

How the signal's distribution changes across scales — is it self-similar, drifting, or scale-free?

Splits the signal into blocks at multiple scales (2, 4, 8, 16, ... blocks), computes the histogram of each block, and measures the Wasserstein (earth mover's) distance between adjacent blocks at each scale. Self-similar signals have constant Wasserstein distance across scales. Non-stationary signals show drift. Scale-free processes follow a power law.

Metrics

w_mean

Mean Wasserstein distance averaged across all scales. Captures the overall level of distributional heterogeneity regardless of scale structure.

w_std

Standard deviation of Wasserstein distances across scales. High values mean the signal's distributional contrast is strongly scale-dependent; low values mean it's roughly constant.

w_fine

Average Wasserstein distance between adjacent blocks at the finest scale. Morse Code (0.394) dominates: its on/off structure creates maximally different local distributions between signal and silence segments. Symbolic Lorenz (0.260) and Sprott-B (0.194) are next — chaotic dynamics with occasional regime switches produce large local distributional variation. Periodic orbits and constants score 0.0 (adjacent blocks have identical distributions).

w_max_ratio

Dynamic range of Wasserstein distances across scales (max/min). Triangle Wave scores 1521: at the finest scale, adjacent blocks have similar distributions (both are local ramps), but at the coarsest scale, one half is ascending and the other descending, creating maximum distributional contrast. Standard Map Mixed (1269) and Logistic Period-3 (1118) have similarly extreme ratios because their periodic structure creates scale-dependent distributional asymmetry.

w_slope

Log-log slope of Wasserstein distance vs. number of blocks. Positive slope means distributional differences grow at finer scales (more blocks = more local contrast between adjacent windows). Negative slope means they grow at coarser scales (fewer, larger blocks reveal structure that fine divisions miss). De Bruijn (6.38) has the steepest positive slope: adjacent small windows differ sharply because the de Bruijn sequence cycles through all byte patterns locally. Devil's Staircase (-0.74) has the most negative slope: its long constant plateaus make adjacent fine-scale windows identical, but coarse-scale blocks straddle plateau boundaries, creating distributional contrast.

Atlas Rankings

w_fine
SourceDomainValue
Morse Codewaveform0.3942
Symbolic Lorenzexotic0.2599
Sprott-Bchaos0.1939
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
w_max_ratio
SourceDomainValue
Triangle Wavewaveform1520.9765
Standard Map K=0.5 (Mixed)chaos1268.6816
Logistic r=3.83 (Period-3 Window)chaos1118.2270
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Logistic r=3.5 (Period-4)chaos0.0000
w_mean
SourceDomainValue
Pulse-Width Modulationwaveform0.1872
Square Wavewaveform0.1483
μ-law Sinewaveform0.1450
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
w_slope
SourceDomainValue
De Bruijn Sequencenumber_theory6.3834
Logistic Edge-of-Chaoschaos3.4461
Fibonacci Wordexotic2.2934
···
Devil's Staircaseexotic-0.7396
fBm (Persistent)noise-0.6030
ETH/BTC Ratiofinancial-0.5976
w_std
SourceDomainValue
Square Wavewaveform0.1674
Clipped Sinewaveform0.1634
μ-law Sinewaveform0.1622
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Logistic r=3.2 (Period-2)chaos0.0000

When It Lights Up

Multi-Scale Wasserstein measures distributional self-similarity — something no entropy metric captures directly. Two signals with the same entropy can have completely different scale profiles: one might have constant distributions at all scales (self-similar), while the other has scale-dependent drift. In the atlas, w_slope separates sources with fine-scale local variation (De Bruijn, Fibonacci: positive slope, adjacent small windows differ) from sources with coarse-scale structure (Devil's Staircase, fBm: negative slope, distributional contrast only emerges in large blocks).

# Quasicrystal Geometries

Open in Atlas
← p-VariationPenrose (Quasicrystal) →