Spectral Analysis

Dominant frequency, spectral slope, bandwidth, periodicity
dynamicalencoding-invariantdim frequency8 metrics

What It Measures

Where the energy lives in frequency space.

Computes the power spectral density via FFT and characterizes its shape. Not just "what's the dominant frequency" but "is the spectrum flat (noise), peaked (periodic), or power-law (fractal)?"

Metrics

spectral_slope

The exponent in P(f) ~ f^beta. Brown noise = -2, pink noise = -1, white noise = 0. Positive slopes (blue spectra) are rare in nature — Thue-Morse (+1.75) and Fibonacci word (+1.63) are the bluest signals in the atlas, their substitution structure concentrating energy at high frequencies. Kilauea tremor (-3.33) and Lotka-Volterra (-3.14) are the reddest: slow dynamics dominate.

spectral_r2

How well does a single power law fit the spectrum? Anti-persistent fBm scores 0.99 (textbook power law). Logistic period-2 scores 0.0 (all energy at one frequency, not a power law at all). This distinguishes genuine 1/f processes from signals that happen to have similar average slope.

spectral_entropy

Shannon entropy of the normalized spectrum. RANDU and dice rolls score 0.95 (nearly flat — energy spread uniformly). Logistic period-2 scores 0.0 (all energy at one frequency). Higher entropy = more frequencies contributing = broader bandwidth signal.

spectral_flatness

Wiener entropy: ratio of geometric to arithmetic mean of the spectrum. 1.0 = perfectly flat (white noise). 0.0 = all power at one frequency (pure tone). Rule 30 scores 0.57 — its pseudorandom output is flatter than most chaos but not as flat as true randomness.

peak_frequency

Where is the maximum power? Stern-Brocot walk peaks at the Nyquist frequency (0.5). Most natural signals peak near DC (0.0). Values near 0.5 indicate the signal's dominant variation is at the shortest timescale — rapid alternation or anti-persistence.

high_freq_ratio

Fraction of total spectral power above the midpoint frequency. Logistic period-2 (1.0) and period-3 (1.0) concentrate all energy at high frequencies. PID Controller and constants score 0.0 (all energy at DC or low frequencies). This is a simpler, more robust version of peak_frequency that measures the balance between slow and fast dynamics.

phase_coherence

Mean coherence of phase across frequency bins. Constants (1.0) and logistic period-2 (1.0) have perfectly coherent phases. Gaussian Noise (0.0005) has no phase structure. High phase coherence means the signal's frequency components maintain a fixed relationship — the hallmark of deterministic or periodic processes. Low coherence means the phases are random, as in noise.

spectral_bandwidth

Standard deviation of the spectral centroid, measuring how spread out the power is around its center of mass. Stern-Brocot Walk (0.21) has the widest bandwidth — its power spreads across all frequencies. Constants and logistic period-2 score 0.0 (all power at a single frequency). This complements spectral_entropy: a signal can have moderate entropy (several peaks) but low bandwidth (peaks clustered near one frequency).

Atlas Rankings

high_freq_ratio
SourceDomainValue
Logistic r=3.2 (Period-2)chaos1.0000
Logistic r=3.83 (Period-3 Window)chaos0.9999
Logistic r=3.5 (Period-4)chaos0.9533
···
Constant 0xFFnoise0.0000
fBm (Persistent)noise0.0000
Lotka-Volterrabio0.0000
peak_frequency
SourceDomainValue
Noisy Period-2chaos0.5000
Collatz Flightsnumber_theory0.5000
Stern-Brocot Walknumber_theory0.5000
···
fBm (Antipersistent)noise0.0001
Brownian Walknoise0.0001
Constant 0xFFnoise0.0001
phase_coherence
SourceDomainValue
Constant 0x00noise1.0000
Logistic r=3.2 (Period-2)chaos0.9998
Logistic r=3.5 (Period-4)chaos0.9991
···
Noisy Sine (SNR 3 dB)waveform0.0005
Gaussian Noisenoise0.0005
Benford's Lawnumber_theory0.0005
spectral_bandwidth
SourceDomainValue
Stern-Brocot Walknumber_theory0.2092
Collatz Flightsnumber_theory0.1720
Continued Fractionsnumber_theory0.1672
···
Constant 0xFFnoise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
fBm (Persistent)noise0.0016
spectral_entropy
SourceDomainValue
Categorical Sensorexotic0.9537
RANDUbinary0.9534
Dice Rollsexotic0.9534
···
Constant 0xFFnoise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
Logistic r=3.5 (Period-4)chaos0.0209
spectral_flatness
SourceDomainValue
Rule 30exotic0.5680
Categorical Sensorexotic0.5656
Dice Rollsexotic0.5640
···
Constant 0xFFnoise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
Logistic Edge-of-Chaoschaos0.0000
spectral_r2
SourceDomainValue
fBm (Antipersistent)noise0.9935
Perlin Noisenoise0.9376
Sine Wavewaveform0.8560
···
Constant 0xFFnoise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
Logistic r=3.5 (Period-4)chaos0.0000
spectral_slope
SourceDomainValue
Thue-Morseexotic1.7999
Fibonacci Wordexotic1.6507
Fibonacci Quasicrystalnumber_theory1.6050
···
Mackey-Glassmedical-3.3298
Lotka-Volterrabio-3.1402
Kilauea Tremorgeophysics-3.0504

When It Lights Up

Spectral slope is the strongest separator between the ordinal view's C1 (red-spectrum oscillators, slope -1.9) and C4 (blue-spectrum chaos, slope +0.3). In the seismic P-wave investigation, spectral metrics were among the top discriminators: earthquake P-waves flatten the ambient spectrum by injecting broadband impulsive energy.

Open in Atlas
← NonstationarityHigher-Order Statistics →