Spectral Graph

Diffusion dimensionality, spectral shape, cluster structure
topologicaldim graph spectrum2 metrics

What It Measures

The effective dimensionality your data's point cloud presents to a diffusing particle -- and whether that dimensionality follows known physical laws.

Delay-embeds the signal into 5 dimensions, builds an epsilon-neighborhood graph on the resulting point cloud, and computes the Laplacian spectrum. Two numbers come out: the spectral dimension (how fast heat spreads on the graph) and the Weyl exponent (how eigenvalues grow with index). For a regular d-dimensional lattice, the spectral dimension equals d and the Weyl exponent equals 2/d. Deviations from these relationships signal anomalous geometry -- fractals, clustered manifolds, or degenerate embeddings.

Metrics

spectral_dim

The effective dimensionality seen by diffusion on the data manifold. Logistic periodic orbits (Period-2, Period-3, Period-4) all score 2.004 -- their delay embeddings land on clean low-dimensional manifolds (circles, figure-eights) where heat diffuses in exactly 2 dimensions. Tidal Gauge (0.10) has nearly zero spectral dimension, meaning diffusion is almost completely blocked -- the point cloud is fragmented or one-dimensional. The gap between 2.0 and 0.1 is the gap between a smooth manifold and a dust-like point set.

weyl_exponent

The eigenvalue growth rate. Wigner Semicircle (2.15) and Circle Map QP (2.01) have exponents near 2, matching the Weyl law prediction for a 1D manifold. Henon-Heiles (1.97) is close behind. Periodic logistic maps score near zero because their spectra have too few distinct eigenvalues for a meaningful power-law fit. High Weyl exponent means the Laplacian spectrum is rich and well-structured.

Atlas Rankings

spectral_dim
SourceDomainValue
Logistic r=3.5 (Period-4)chaos2.0045
Logistic r=3.2 (Period-2)chaos2.0044
Logistic r=3.83 (Period-3 Window)chaos2.0039
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Tidal Gauge (SF)geophysics0.1034
weyl_exponent
SourceDomainValue
Wigner Semicirclequantum2.1480
Circle Map Quasiperiodicchaos2.0066
Hénon-Heileschaos1.9701
···
Constant 0xFFnoise0.0000
Constant 0x00noise0.0000
Logistic r=3.5 (Period-4)chaos0.0000

When It Lights Up

Spectral Graph is the atlas's cleanest test for manifold dimensionality. It separates smooth delay embeddings (spectral_dim near 2.0) from fragmented point clouds (near 0) with almost no overlap. The spectral_dim and weyl_exponent pair together provide a two-parameter characterization that other topological geometries lack -- Persistent Homology counts features, Cayley measures graph growth, but Spectral Graph directly measures how space "feels" to diffusion.

Open in Atlas
← Mostow RigidityPersistent Homology →