Hyperbolic (Poincaré)

Hierarchy depth, branching, boundary clustering
topologicaldim 25 metrics

What It Measures

How much your signal's structure looks like a tree -- branching, hierarchical, pushing toward the boundary of a curved disk.

Maps consecutive byte pairs into the Poincaré disk, a model of hyperbolic space where distances grow exponentially near the edge. Flat, uniform data spreads evenly across the disk. Hierarchical or heavy-tailed data gets shoved toward the boundary, because the disk has exponentially more room out there -- the same reason tree-like data embeds naturally in hyperbolic space. The geometry computes where the mass centroid sits, how far points are from the origin, and how far apart they are from each other under the curved metric.

Metrics

mean_hyperbolic_radius

Average distance of embedded points from the disk's origin. L-System Dragon (4.72), Pulse-Width Mod (4.72), and Morse Code (4.72) all push points to the boundary, where hyperbolic distances explode. Bearing Inner (0.22) stays close to the center. This metric separates signals with extreme byte-value concentrations from those with more centered distributions.

curvature_structure

Ratio of local curvature estimates at different scales in the Poincaré disk embedding. Baker Map (200.3), Tent Map (198.1), and Logistic Chaos (193.1) score highest — their space-filling dynamics create multi-scale curvature structure. DNA sequences score near 0.7 (almost flat curvature — their delay embeddings are geometrically simple). This measures how much the hyperbolic curvature varies across the point cloud.

knn_scale_ratio

Ratio of k-nearest-neighbor distances at two different k values, measuring how fast the local density changes with scale. fBm Persistent (46.4) and Sawtooth (45.9) score highest — their smooth dynamics create point clouds where density changes dramatically with scale. L-System Dragon and Rule 30 score 1.0 (uniform density at all scales). High ratios indicate multi-scale clustering; ratio near 1.0 indicates uniform local structure.

spatio_temporal_corr

Correlation between spatial proximity in the Poincaré disk and temporal proximity in the original signal. NYSE (0.75), Nikkei (0.74), and NASDAQ (0.74) score highest — temporally nearby points are also spatially close in hyperbolic space, reflecting persistent dynamics. Pomeau-Manneville (-0.55) scores most negative — temporal neighbors are spatially distant, the signature of intermittent switching between distant states.

temporal_variance

Variance of the hyperbolic radius over time. Collatz Parity (18.9), Symbolic Henon (18.2), and Symbolic Lorenz (17.8) score highest — their binary dynamics create alternating near-center and near-boundary points. Constants and logistic period-2 score 0.0 (fixed radius). This captures how much the signal's "hierarchical depth" fluctuates over time.

Atlas Rankings

curvature_structure
SourceDomainValue
Baker Mapchaos200.3898
Tent Mapchaos198.0556
Logistic Chaoschaos193.1145
···
DNA Thermusbio0.6526
DNA Phage Lambdabio0.7438
Codon Usagebio0.7582
knn_scale_ratio
SourceDomainValue
fBm (Persistent)noise46.3661
Sawtooth Wavewaveform45.9325
Chua's Circuitexotic41.9344
···
Constant 0xFFnoise1.0000
Logistic r=3.5 (Period-4)chaos1.0000
Logistic r=3.83 (Period-3 Window)chaos1.0000
mean_hyperbolic_radius
SourceDomainValue
Pulse-Width Modulationwaveform4.7174
Morse Codewaveform4.7174
Square Wavewaveform4.7174
···
Bearing Innerbearing0.2166
Ambient Microseismgeophysics0.2621
BTC Returnsfinancial0.2882
spatio_temporal_corr
SourceDomainValue
NYSE Returnsfinancial0.7481
Nikkei Returnsfinancial0.7440
NASDAQ Returnsfinancial0.7389
···
Pomeau-Mannevillechaos-0.5557
DNA Thermusbio-0.4459
Collatz Gap Lengthsnumber_theory-0.4127
temporal_variance
SourceDomainValue
Collatz Paritynumber_theory18.9112
Symbolic Henonexotic18.2333
Symbolic Lorenzexotic17.7583
···
Constant 0xFFnoise0.0000
Logistic r=3.2 (Period-2)chaos0.0000
Logistic r=3.5 (Period-4)chaos0.0000

When It Lights Up

Hyperbolic geometry is most useful for separating hierarchical data from flat or periodic data. The spatio_temporal_corr metric uniquely identifies signals where temporal and geometric proximity coincide — financial returns cluster tightly, while intermittent dynamics show negative correlation. The curvature_structure metric provides a multi-scale geometric fingerprint that separates space-filling chaos from low-dimensional manifolds.

Open in Atlas
← Persistent HomologyVisibility Graph →