Work

Toward a Structural Curvature Framework for Spatial Statistics: A Holonomy-Inspired Approach to Boundary-Sensitive Anomaly Detection

Year: 2026
Type: preprint
Abstract: Spatial statistics and environmental monitoring frequently hinge upon the precise detection of structural boundaries, ecological interfaces, and abrupt transitions within heterogeneous landscapes. However, conventional spatial indices (e.g., Moran’s I, LISA) anddifferential image processing filters suffer from a fundamental limitation: they measuremere variations in amplitude or local variance rather than structural discontinuities, rendering them highly susceptible to smooth environmental gradients and micro-scale noise.Furthermore, existing symmetric positive-definite (SPD) manifold methods primarily evaluate pairwise geodesic distances, failing to distinguish coherent regional drift from multidirectional structural collisions. To address this theoretical gap, we propose a geometricframework—Toward a Structural Curvature Framework for Spatial Statistics—byreformulating multivariate spatial observations (d ≥ 2) via SPD manifold geometry andlog-Euclidean tangent mapping. Drawing inspiration from lattice gauge theory and discretedifferential geometry, we construct a discrete holonomy operator along elementary plaquettes, whose generator Fp serves as a curvature-inspired measure of incompatibility amonglocal covariance geometries rather than a detector of intensity discontinuities. We establish foundational properties: Lemma 1 derives an explicit, dimension-free upper bound∥Rp∥op ≤ Cδ3 for the local Baker-Campbell-Hausdorff (BCH) remainder via a scalar majorant series; Proposition 1 and Proposition 2 establish immunity to commutative gradients, feature-space gauge equivariance of Fp, and orthogonal invariance of the scalar scoreZp =∥Fp∥F; and Proposition 3 (Interior Noise Decay) characterizes how the expectedscore E[Zp] decays as O(α2/n) outside a w-neighborhood of structural interfaces, complementing empirically observed boundary localization. Validation across macroscopic boundary benchmarks (Case F), commutative gradient controls (Case C), pure Wishart null fields(Case F”), and high-frequency checkerboard stress tests (Case F’), alongside an explicit characterization of known failure and model-collapse modes, establishes the operational scope ofthe proposed framework.
Author: Yusei, Inaba
Language: English
Open Access status: green
OpenAlex ID: W7236628241
Cites:
Cited by:
Related to: 9