Hölder Regularity

Local roughness, regularity spectrum, singularity strength
scaleencoding-invariantdim function space4 metrics

What It Measures

How rough or smooth the signal is at each point — and how much that roughness varies.

If you zoom into a smooth signal, it looks linear. If you zoom into a rough signal, it stays jagged no matter how far you zoom. The Holder exponent quantifies this: high = smooth, low = rough. This geometry computes it at every point and asks: is the roughness uniform (monofractal) or does it vary wildly from place to place (multifractal)?

Metrics

hurst_exponent

The global roughness summary. Sine waves score 0.92 (very smooth, persistent). White noise scores ~0 (uncorrelated). Logistic maps in the period-doubling cascade score -2.0 (actively anti-persistent — each value fights the previous one). In the seismic P-wave investigation, earthquake arrivals scored dramatically lower than ambient noise (d = 9.06): P-waves are impulsive, ambient microseisms are smooth.

holder_mean

Average local regularity. Wigner semicircle (0.99) and triangle wave (0.98) are the smoothest signals in the atlas. Logistic period-2 (-3.2) is the roughest — it alternates between two values with no interpolation.

holder_std

How much does roughness vary? English literature scores highest (1.87): some passages are smooth (common words), others are jagged (rare words, punctuation). Fibonacci word scores 0.0 (perfectly uniform roughness at every point — it's monofractal).

holder_min

Minimum local Hölder exponent. Rudin-Shapiro (0.66) and Triangle Wave (0.56) have the smoothest worst-case points. Logistic period-2 (-3.2) and full Logistic Chaos (-2.0) have extremely rough minima. The smaller the minimum, the rougher the signal's worst-case point — the local analog of multifractal width.

alpha_autocorrelation

Temporal persistence of the Hölder exponent sequence. Hodgkin-Huxley (0.77) and Van der Pol (0.76) score highest — their smooth oscillatory dynamics create slowly-varying local regularity. Number-theoretic sequences (Divisor Count, Prime Gaps) and anti-persistent fBm score 0.0 (roughness varies randomly from point to point).

increment_autocorrelation

Lag-1 autocorrelation of the increment (first-difference) sequence. Clipped Sine (0.98) and Lotka-Volterra (0.98) score highest — their smooth waveforms have highly predictable increments. Circle Map QP and Critical Circle Map score 0.0 (increments change unpredictably).

Atlas Rankings

alpha_autocorrelation
SourceOriginValue
μ-law Sinesignal-synthesis0.6014
Logistic Chaositerated-maps0.4235
Heisenberg Walksymbolic-dynamics0.3738
···
Seismic Noise (ANMO)geophysical0.0000
x86-64 Machine Codealgorithmic-bytes0.0000
Accel Joghuman-activity0.0000
holder_mean
SourceOriginValue
Gaussian Collatz Orbitnumber-theory1.3449
Spike Trainstochastic-process1.0699
BTC Volatilityfinancial1.0224
···
Intermittency Type-IIIiterated-maps-1.9619
Logistic r=3.5 (Period-4)iterated-maps-1.8938
Heisenberg Walksymbolic-dynamics-1.8842
holder_min
SourceOriginValue
Rudin-Shapirosymbolic-dynamics1.0000
Sawtooth Wavesignal-synthesis1.0000
Tank Drain Cascadeself-organized-criticality0.9964
···
Solar Wind IMFastrophysical-2.0000
x86-64 Machine Codealgorithmic-bytes-2.0000
Henon Mapiterated-maps-2.0000
holder_std
SourceOriginValue
Goldbach r(2n)number-theory1.9736
ECG Beat Conformityphysiological1.9109
Beta Noisestochastic-process1.8885
···
Collatz Cycle Wordnumber-theory0.0000
Pell Wordsymbolic-dynamics0.0000
Penrose Substitutionsymbolic-dynamics0.0000

When It Lights Up

Holder Regularity was the #2 discriminator in the seismic P-wave investigation (Cohen's d = 9.06), detecting earthquake arrivals through the collapse of local smoothness. It separates the Distributional view's C1 (smooth oscillators) from C4 (anti-persistent chaos) along the persistence axis — the biggest gap in ordinal space.

Open in Atlas
← Ulam Spiral (Sacks)Multi-Scale Wasserstein →